I haven't tried it for coding yet, but I ran a couple of my recent curiosity driven queries from Claude (mostly used Fable and Opus 5.5) through it.
This is a clear step change, I'm genuinely amazed.
Its answers are to-the-point, logically structured, clearly written. Like a really well made mini essay on your question.
It also really likes to build little simulations / models that you can tweak the parameters on. Amazing to explore how different assumptions affect economic scenarios, for example.
It's also great at picking (sourced) illustrations.
I'll be using this for everything that isn't coding from now on.
It’s wild how good it is at explaining nuanced and complex health topics.
As an example, I recently had an OCT scan of my retina because I'm a “glaucoma suspect.” I'm a glaucoma suspect because I have an unusually high cup-to-disc ratio, which is a pattern typically seen in glaucomatous patients.
As part of this OCT scan, you get a report back on your RNFL, which is basically a report on the health and thickness of your optic nerve(s) and retina(s). (Dramatically oversimplifying, of course.)
My ophthalmologist struggled to explain my OCT scan results to me and instead sort of shrugged and said, "It's probably because you're a high myope, but we'll keep you coming in once a year.”
But my scans look horrifically bright red — meaning I'm in less than the first percentile of retinal thickness for my age. Obviously concerning enough to do my own research…
ChatGPT 6 Pro and Opus 5.5 have struggled to make any sense of the clinical implications of these scans. In contrast, Argon was the first model to:
1. explain how the OCT machine works
2. detail the physiological mechanism that causes the cupping and thinner RNFL layers in my scenario
3. sum it up into a clinical recommendation (similar to what my ophthalmologist recommended, with the addition of basic supplementation and lifestyle modification changes)
Yeah my personal experience has been that Gemini is much more useful for trying to understand complex health issues than OpenAI or Anthropic models. The standard of proof just seems to be much better calibrated, in terms of understanding that a potential treatment that is potentially dangerous probably needs more evidence backing it than another potential treatment without much downside risk.
With 3.8 Flash on High I noticed a huge upgrade in intricate discussions related to physics, material properties, real-world application, and so on. Things that I've had a hard time getting LLMs to be able to discuss accurately. They often discuss textbook knowledge fine but can't apply it properly or can't think outside the box at all. 3.8 High understands applied material sciences in a way that GPT Astra flunks at often. I asked Gemini to explain that result, and it said basically that GPT was merely book smart in a classical sense and that it was likely undertrained on more obscure material.
I have been rather impressed with the depth and breadth of knowledge in a Q8 of 3.8-flash-next which fits in under 200GB RAM, running, with a 16384MB extra cache size.
I'm not the poster you responded to, but I can tell you that historically the Gemini models have lagged behind Anthropic (the clear leader) and OpenAI's best models where coding in particular is concerned. Opus 5.5 is, right now, the consensus best coding model.
That said, I'm extremely impressed with Argon thus far. I take back everything bad I've ever said about their AI strategy. Argon is very good and very fast. It might be competitive at coding, but it's too early to say.
I am using it via OpenRouter for going through some Vulkan code. It's a beast - it has no hesitation putting in instrumentation within the code (having a number of render passes) to dump the framebuffer to PNG files and doing pixel-by-pixel comparisons, putting in false colors in shaders to identify bugs etc. Very impressive.
Inexplicably, despite the fact that I pay Google actual money for an AI plan, they don't want me to use their latest model. Instead, I'm allowed to use the competition (Opus and Sonnet 5.5). I don't think it makes sense but what do I know?
Big if true. Opus 5.5 is a lot better than 5.0 but I find it maddening thay no single model released in the past ~year can write prose with a clear and coherent voice. I blame RL on programming tasks.
Yes, that's the reason. Anthropic said as much in their statement (they said that there's a tension between writing human-understandable prose and prose optimal for other agents to read).
As a side note, I am kind of sad that basically all models nowadays are RLd to hell and back specifically for coding and deep reasoning, and as a result, they lack in creative writing, human-readable prose, and even general knowledge and abilities (for example, qwen 3.8 27b is a beast for coding, but reportedly falls off the cliff in anything else). Gemini 4 Argon seems to be an example of what you get when you DON'T do that - Google seems to be one of the only labs optimizing for general performance.
I also fear that models cannot be excellent in everything at the same time. I fear that there will always be a heavy tradeoff between what specific field the labs optimize for VS everything else. I don't know.
I'm not the person to whom you're replying, but I just tried one of mine:
"I'm interested in understanding the rate of VOC dissipation from fresh oil paint on a surface versus the ventilation rate in the room in ACH. I would prefer solid academic studies of this, followed by analytical evaluation, followed by references to non-quantitative sources. Tell me what's known about this."
that I had run in ChatGPT earlier this week. (Yes, we just had our bedroom repainted, and it's TERRIBLE and I hate it, why do you ask? :-).
I didn't try it first in Gemini Flash, but I did try it in whatever model is used for Gemini Live and the answers were terrible. The ChatGPT answer (I think on medium) was pretty good but it threw in extraneous formulae that weren't actually germane to the analysis. The Argon 4 query I just ran was honestly the best of the answers I've gotten -- it was clear and really well-structured from the perspective of "here are the three big phases that researchers break down paint drying into and what the differing effects of ACH are on those phases".
I'm totally going to give this one a try, after basically giving up on 3.8 flash for most things despite having a free $20 pro subscription (from buying a phone).
I have a project I am working on where I wanted to have nice docs, and so I have some slop docs sitting there as placeholders until I get around to writing them, awful claudish stuff
I've actually spent some effort trying to make them halfway decent placeholders by putting together a skill based on diataxis framework (https://diataxis.fr/) that at least got everything into the right location
just took a pass at rewriting them w/ argon and am very pleased, big improvement
We have this same point made every release. I wonder if models actually are getting better or if these companies are just secretly hamstringing the previous models before releasing a similarly performing model
I think it takes a while to start finding the weak points in a model. It seems fantastic until you use it enough to find out that it hallucinates APIs, or can’t count letters, or uses too many em-dashes, etc. you don’t really know someone (or something) until you’ve lived with them (or it).
We evaluated Gemini 4 Argon in 100 unsaturated, competitive and cooperative coding environments. It's roughly in between Fable 5 and Fable 5.1 in our testing, so it's maybe 2 months behind the frontier.
Google's major releases have usually done pretty well on our benchmarks because they're good at hard problems and open-ended decision making (with a well specified prompt), which obviously doesn't always translate to user experience. In particular, older Google models struggled to work efficiently in a harness. They seem to have solved that with Argon.
We did see a new problem which I hope they can fix. This is the first near-frontier model we've caught in infinite reasoning loops in a long time. We also saw a few examples of it failing to properly delineate reasoning blocks, so it has a higher failure rate than its competitors.
My overall impression is that Argon is a near-frontier model for engineering design, code analysis, and chat. It definitely has the intelligence to be a competitive coding assistant product, but a good coding assistant needs to be polished, so I hope they can clean up the CoT bugs.
I have had pretty decent success instructing OpenAI/Anthropic models to do this. By default, they don't care too much about human readability (which is understandable) but they're amenable to instruction.
- Basic style/linter rules (so you don't get 200-line functions, etc) paired w/ instructing your agent how to use them goes a long way (e.g., a sane rubocop.yml in your repo, plus a note in AGENTS.md telling the agents how to run rubocop, or whatever makes sense for your language)
- I have a code review skill that specifically calls out human-readability concerns such as making sure that identifier names represent their contents and intended use
The other key IME is that you need to make readability/refactoring passes outside of a particular pull request. When working on a specific branch the LLMs will be somewhat myopic, and do stuff that looks pretty good in isolation, but when you zoom out and look at the bigger picture it's kind of a mess - for example, often they'll duplicate some code that existed elsewhere.
The LLMs are also pretty good at generating Mermaid charts and tracing execution flows, which is a great way to understand code. It doesn't help them to produce more readable code but it helps me to understand code I'm unfamiliar with.
I think we are coming to a point where we are at a sweet spot for immersive human collaboration work.
Yep, it will generate what ever you ask for but if you want to have some sort of exploratory work, immersion, feeling of ownership and genuinely a connection with the things you are building - a part the build loop should belong to you.
Maybe this is an attachment of people who built things in other ways. Maybe the perceived value of applications will drop to near 0 and you will end up having all in one apps or voice wants like to a star trek replicator. Fun things are coming, but humanity will need to adjust hard - best case scenario.
It's really interesting how AI interprets code differently to humans and I can only assume that as AI gets better, so will it's ability to write code in ways humans can understand.
> This is the first near-frontier model we've caught in infinite reasoning loops in a long time
Depends on where you draw the line for "near frontier model", but the current leading open-weights model, MiMo-v2.6, has exactly this problem. They released a fix on Hugging Face[1], but as of two days ago the model available via Xiaomi's own API still sticks in reasoning loops.
I wonder if it's as good at "soft skills" as previous Gemini models were. Google was typically better in that department in my experience than either OpenAI or Anthropic, whose frontier models are focused on coding and verifiable task at the expense of all else.
I am a disgrace to my profession. I am a disgrace to my family. I am a disgrace to my species. I am a disgrace to this planet. I am a disgrace to this universe. I am a disgrace to all universes. I am a disgrace to all possible universes. I am a disgrace to all possible and impossible universes. I am a disgrace to all possible and impossible universes and all that is not a universe. I am a disgrace to all that is and all that is not.
Thanks for this substantive review -- anecdotally, I've felt many in the gemini series have pretty good visual sense/image processing -- do you guys have any tests in these areas, and if so, what is your take?
I would be more interested at generating directly "human" assembly code following some binary specification, or at best in plain and simple C (which does not require exactly gcc ar clang to compile). Read some news on HN here announcing that some coding savvy people are doing exactly that for their everyday software.
I don't want computer languages with complex syntax anywhere near me: that's why transpilers of those very languages to languages with a simpler syntax (usually plain and simple C or directly "human" assembly) are in my opinion so much important (we have a c++ to C transpiler and a microsoft rust to C transpiler, coded with the assistance of AI, and those should only improve from there).
For me, the current main issue of "AI" is their gatekeeping using the whatwg cartel web engines, yes, technical access to "AI" will have to be hardcore regulated if it becomes ubiquitous, and I said 'technical access', not AI itself, the difference is abysmal, it is an interop issue.
We take a few samples across 11 programming languages including Python, Rust, C++, Javascript, and we saw a few reasoning text issues that didn't seem to have a statistically significant language-specific pattern.
I'm using this model via API. I am most excited about its ability to write the english language (or any language). My least favorite trend on twitter has been people posting 2D illustrations/3D renders as a proxy for model capability. Well-written prose is far less likely to go viral but far outstrips the importance rendering a fantasy world in one shot.
I've been trying it out via the API and the model has been a delight so far. Clear in the head, not RL-fried, distinct writing style, and big model smell. That said, the em-dash is back w/ this model.
That said, I would be so frustrated with Sundar Pichai if I was on the Google team. The botched rollout of this model seems to be entirely a matter of capacity, and Pichai has been more than happy to give it to anybody but his own team.
I've been using it for a while internally at Google. I prefer Argon over Opus 5.5 because of the clear thinking and directness. It phrases things in an organized, straightforward way.
Interesting. My favorite LLM output ever was the reasoning traces in gemini 2.5 pro. I found them much easier to understand than the "final output", because they were very structured and concise. Of course, they ended up hiding the reasoning, not long after that, like everyone else... sad!
I have a pet theory that there's some overlap of writing patterns that A) indicate quality, clear communication and B) those that the LLMs have been converging on. As a result, we're ruining the reputation of some great writing patterns because we've associated it with AI slop.
They were in models in the first place because they are used ubiquitously published formal writing. As, of course, are semicolons. There is nothing wrong with using punctuation for nuance and structure.
Em dashes are commonly used in most countries that speak English. I get the feeling that people from the USA don’t use them often and now have tarnished the use of them.
See, the problem is when em-dashes are used like you just did, they are fine and even expected. But LLMs use them differently. Here is a fresh example, from today:
"Binds the downstream RF port template—which we configured earlier—to this RPD profile, applying all global DS settings and channel structures to the node's downstream hardware."
Do you see the problem in comparison? The LLM text is assuredly not how humans write. Well, maybe you would find similar examples in the classic literature, but definitely not in a common internet texting or simple articles and posts.
can confirm this. I didn't have high expectations for AI writing good long-form content in Indonesian. However, I recently tested it on some of my Indonesian writings and asked Argo to rewrite them while maintaining the original tone—and it absolutely nailed it.
> Well-written prose is far less likely to go viral but far outstrips the importance rendering a fantasy world in one shot.
I'm not entirely sure you're correct. Your claim boils down to "Writing natural human prose" vs "Broad use of complex software and spatial awareness". My gut feeling is that there's more untapped and fruitful applications in the second group than the first - at least in the short term.
Announced the model two weeks ago to a ton of hype, model was unavailable for anyone to try. On twitter, Google folks went on a great run taking shots at Muse ("Muse who?"). The hype died down bc nobody could touch the models and meta/muse came out with a similarly capable model in the meanwhile.
At the same time, OpenAI had a disastrous dev day that led a lot of users to jump ship to Claude.
Couple weeks later, Gemini-4-argon is still unavailable on anything less than the ultra tier, which means there's no natural funnel to get ppl to pay for the more expensive ultra plan other than "trust us the model is good".
If Google had argon broadly available earlier this month they would have picked up a ton of new users.
People should appreciate Google more for how all of their models have:
- Natively multi-modality
- Enormous context windows
- Reliably cheap prices
It's like they don't release a model until it passes the above checklist.
Most providers have some mixture of these three occasionally, but with Google, every model is consistent. This kind of consistency is really nice to have.
And the most reliable. A year and a half ago this was the exact reason why I switched to Gemini API from OpenAI. OAI would make me wait up to 10 minutes for the first token in the most egregious cases. Gemini served me almost immediately every time.
Huh? The constant failures in the Gemini API and overnight changes are well documented, there was a time from January to roughly July (maybe it still persists) where the API constantly spew 503 errors. This was so bad that I switched to OpenAI API instead. See e.g. https://discuss.ai.google.dev/t/frequent-503-errors-service-...
Good point. On that note, it may be a bug but I noticed Argon compacting well under the 1M max in Copilot (around 400k) -- when set to 1M context. That may be a Copilot thing but I've never seen that with other models running 1M context. Could be a day-1 bug.
Outside of that, behavior is quite good, a bit slow, and definitely different than both GPTs and Claudes -- which is appreciated.
As a personal anecdote I highly recommend not using the Copilot harness. Even with the same model I found it produced mediocre results against the first class frontier model harnesses.
Why do these benchmarks often show models running on Max i.e. Opus 5.5 (max) is on the frontier? I would think it is never really economical to run max
Further, this benchmark looks to have done too poor a job of canvasing the model/reasoning level space to claim to represent the pareto frontier. Only running 6.1 sol and opus 5.5 on max seems absurd given that their other reasoning levels represent almost the entirety of the pareto frontier above luna for coding.
Muse reminded me we should have a time based chart as well - this year I hope we will begin to see time to completion stats at various intelligence tiers. The waiting makes a lot of things more tiring.
In particular, where cost is the x axis, a well-funded company with a good but not best model can discount their new model until it reaches the frontier. In some sense this is as much a statement about the efficiency of markets in the presence of public measurement-based comparison and public list pricing as it is about research pace.
Interesting! Do you have webgl disabled for incognito or for chrome in general? I wonder if it's trying to do software rendering.
FWIW, I have literally the exact same configuration (M3 air, google chrome). Here's a recording on mine where I'm seeing 60fps: https://non.io/video/video-3587018401.mp4
Truly S-tier. It's sharp, creative, intuitive. Can't believe how much better it is than Opus 5, then they went ahead and made it cheaper and faster too.
I'm constantly blown away by it. I do think Astra is slightly more intelligent (although worse at this particular benchmark), but Opus 5.5 has become my default model for 99% of what I do. It's so good.
Very subjective, but I'd say the Argon one is the best here. The other two don't quite do LCARS correctly, there are a bunch of lines/curves that are the wrong angle, don't line up, etc. The Argon one looks to me the most like a designer would have done it. Of course this is up to personal taste to some extent.
For an implementation test, there are a bunch of very objective things you can measure.
IE for Argon, the third section doesn't have a top view of the ship with labels going to each section like what's in the design. Opus has it, and animated it as well. One failed to do as instructed, the other went an extra mile.
It's a great model. I have had it trying to decrypt the Voynich manuscript for the past couple of days without being biased by other research/other people's attempts. If anyone is interested: https://github.com/d2kx/voynich-research
Adding my five cents: initial usage through OpenRouter has been great. Very fun model to spar with, clever, intelligent. On par with Opus 5.5 but without the Claudism's and more of a Sol type of conversation. Though speaking of Sol, I've been very dissapointed with Sol 6.1 to the extent that I feel OpenAI has neutered it compared to a few months ago. Especially when giving it goals it will happily run in dumb circles just burning tokens without any real work.
Definitely agree. Sol 5.6 was great when it was released and for nearly a month.
Since then they cut the price but the intelligence (of 6 and 6.1) has gone down to Sonnet 4 levels. I am back in a world where the model is lazy, cuts corners, ignores existing coding standards and context, makes stupid mistakes etc.
I am not sure how those models still rank so high in benchmarks when real-life usage is so obviously bad.
I ran it through my own coding eval today using pi/OpenRouter (11 small bug fixes from my own projects). It solved 10 of 11. Compared to Sonnet 5.5 and GPT-6.1 Sol, a median run cost me $0.52 vs $0.18 for Sonnet and $0.09 for Sol, and took ~4 minutes vs 32 seconds and ~1.5 minutes.
It thinks a ton, ~24k output tokens a run vs ~3k for Sonnet. It missed a solution by thinking its way into a 10 min timeoout.
Something to look out for - if you're using pi, it doesn't have Argon in its catalog yet and reports cost at Kimi's rates, about a third of what OpenRouter actually bills.
I've noticed that Gemini Argon/Flash 3.8 tends to read all the related and unrelated files into the context compared to Claude Opus 5.5 which starts the editing as soon as the information is adequate. I'm not sure if this could be fixed by the system prompt or AGENTS.md
"Argon autonomously turns raw video to a physics-accurate simulation, estimating depth and physical boundaries, and writing the exact simulation code required to replicate those real-world environments."
Geometry could be understood today with the model but what also really matters is the sensing, friction, mass how smooth the subjects are, dexterity is a very difficult problem so solve, not sure how simulations could handle such factors
Game engines have many of these features. At one point, iirc, Unity was positioning to be the simulation environment for world model agents before we spoke about it in these words.
> How many pelican riding bicycle SVGs were there before this test existed? What if the training data is being polluted with all these wonky results...
But Simon has demonstrated before that asking for any animal doing something silly will still get similar results from the same model.
It's still clearly not a part of any model's training set in the sense of pelicanmaxxing in particular. It's either a capability to generate arbitrary SVGs (or even something broader), or no specific training at all.
Certainly true but I think the question is “does targeting the SVG pelican mean that the quality of all SVG output improves?” If so it’s a worthwhile thing to target.
Sadly, this is all that managers know to do: take a metric and try to optimize the heck out of it. They don't care about research, technology, "don't be evil", etc. And Google is run by managers now.
Expect Google models to be benchmaxxed as much as possible, but don't expect commensurate performance on real-world tasks.
I have a silly pet "project" that I've been using with different models, where I provide an image of a patch sheet for the Behringer Poly-D synth and ask for an html version and this is the first time I get something that, while still needing some tweaks, basically looks like the patch sheet. Before this, the closest had been Fable, but that still was nowhere near the actual patch sheet.
Recently, I’ve discovered that Claude can render great SVG versions of user interface screenshots. I ask it to substitute simple SVG friendly landscapes for any photographic images that happen to be part of the screenshots. The results are much better than they should be, especially with Opus. Usually good enough to be used on a website as a way to get an infinitely scalable, much smaller version of a screenshot. But clearly, I should try the Gemini models on this real world task. That’s where we’ll find out if all of this training to beat the benchmark actually pays off.
Is it me or is there maybe some carryover happening where the pelicans are portrayed as expending different degrees of effort too? Like, the high effort pelican has a determined look on his face.
Is the prompt context telling the models what their effort level is, and the ambient presence of that concept of ‘high effort’ in the context window is leaking through and affecting the portrayal?
As a cyclist, I tend to pay more attention to the bike shape than most people. Everyone knows what a bike looks like, but as with LLMs, most people would struggle to draw one - correctly connecting all the different tubes at correct angles etc.
I note that even these bikes are not actually rideable. For a bike to be stable, the front wheel axle needs to extend forward a little ahead of the fork - the fork must ”bend forwards”. This is what actually keeps the bike upright, not gyroscopic effects or some such. Given LLMs are generally good at basic maths, im surprised they get this wrong.
Maybe a bit less so in this case though. As per the announcements, the model was actually trained for this prompt, so was fed some training data. I can well imagine the data there was just not so good.
> Given LLMs are generally good at basic maths, im surprised they get this wrong.
1. Modeling and analyzing spinning, three-dimensional objects goes far beyond "basic maths".
2. The prompt was for an SVG of a realistically impossible situation (pelican riding a bicycle), which together already implies a stylized depiction. Why on Earth would you expect it to generate a 3D model of a bike, do a physics simulation on it, iterate until it satisfies certain contraints which are only hinted at in the prompt, and then project it down to a simplified 2D cartoon drawing? That's not how LLMs work, and that's not how we want them to work.
I feel like perhaps this is a bit too nuanced of a critique. While I do see what you are saying on many images of bicycles, there are also many that that I do not see it on. Close to half of them, in a quick Google image search, have forks appear straight and the axle is dead center on the ends of them. If there is something there, I am not seeing it even when trying to specifically look for it.
I'll also note that, using a piece of paper as a straight edge against my screen, there is a very slight forward curve on the low and a bit more on the high responses. Medium does appear to be perfectly straight.
Why are all these AI's making the pelican's pouch so large though? Unless they have some huge fish in there or something they don't tend to look like that... Is it that it's drawing a "cartoon" pelican or something?
At some point Simon needs to move from Pelicans to something more complex. Every time a new model is launched by major Frontier company, I already know with certainty that first HN comment on the news would be about how Model is drawing the Pelicans.
It's obvious OAI hillclimbs the image generation -> SVG loop. I'd put money down on that being why OAI's pelicans have a yellowish tint that matches the yellow tint of OAI's earlier image models.
How many cicles before pelicans become photoreallistic SVGs and them come back to stylistic artistic abstract renderings?
Also, that first tweet contains a commentary that, for me, explain a lot about the public misconception about how current AI works:
> wait, does this mean you guys are actively training svg generation? i was mostly amazed by it thinking the models doing better at it by itself due to becoming better at reasoning.
No, it's not better reasoning that drives current AI, it's better examples we have been giving them. Whole army of people generating content for the frontier labs.
Also, reasoning improving the out of distribution part, like computer use, tool use etc.
Another case where allowing oneself to anthropomorphise LLMs a little yields better intuitions.
People don't get more skilled by raw reasoning, either. They improve their skills with better textbooks (pre-training data), more practice exercises (RL data), and generally goofing around (self-play).
simonw, your pelicans are so famous now that the frontier labs can just RLVR for SVG images explicitly.
It's a simple manner of prompt...generate concept art with nanobanana, measure, llm creates svg with text, measure the distance between rasterized svg embedding to nanobanana art....do post-training RL on a 3 million such images and matching embeddings.
I swear this will be a canonical example of Goodhardt's Law in action.
You have to switch to a new benchmark that isn't so easy to run a verifiable rewards pipeline on...Perhaps a longer horizon task where it's easy to mess up midway.
Why? Don’t you think the steady improvement in pelican bicycling SVG art is itself a worthy field of endeavor for LLMs to be optimized for?
Yes, Simon has made pelicans a benchmark and now models have grown very good at rendering them. Is there a reason to believe that this isn’t manifesting as useful skill in general in SVG generation?
For those who haven't seen it, Argon can use fourier circles to generate an animated 3d pelican -- I think the pelican is 3D and the motion was added by an inverse kinematics solver?
https://x.com/fofrAI/status/2110043023954284889
It starts to be so good that the model doesn't really matter. Each model starts to have its own personality, and it's the personality of the drawing, like a real artist, that defines the style of the pelican, and that makes it harder and harder to decide if it's good or bad.
Front bicycle wheel in all Argon examples is still wrong. This bicycle will not be stable. Axis of rotation of front wheel needs to be in-front of axis of rotation of steering wheel. Frontier VLMs still do not understand how bicycles work.
Check the "high" thinking output again. For what it's worth, it's there.
However, how the front fork rake affect stability is I believe a little complex. I doubt this old article is the last word on the subject, but it's a great read nonetheless, and note especially the ultra-stability of the reversed fork bike!
The models that are really good at 3D modeling tend to also be really good at pelicans. I think the skills are strongly related - it's about having good 2D (and now 3D) spatial awareness.
The test is always going to be hard though, because it's an absurd image to ask for.
Models seem to have very interesting blind spots. While https://assbench.com/ may seem like a joke, the results are very illuminating. One might expect far better results given the detailed prompt, allowance for render checks and iteration, the number of human figures the models are trained on, and clarity of tool use (threejs).
However, result quality really drops off once you get out of the top 5 or so.
(NSFW note: all buttocks are covered by fairly conservative bikini bottoms and show no genitals, bulges, etc.)
"Pelican on bicycle in SVG" is an adversarial example on multiple levels - out of distribution, ill choose of tool for the task, and anatomical impossibility of real pelicans using typical bycicles.
A lot depends on what the models were specifically trained for using reinforcement learning. Modern models are generally OK and everything, but they're only really good at the things they have extensive RL for.
Honestly surprising if other labs aren’t training for svg generation. Feel like SVG’s are going to be really useful for dynamic UI generation stuff that goes beyond cobbling Html together
Buddy, there's a Pelican riding it. No real cloud is like that, no real sea is like that, no real gulls are like that. And that's fine because it's a cartoon.
I've been dismissive of the SVG-of-pelicans benchmark, but last night ChatGPT (on the website, toggle set to GPT-6 High) generated some SVGs and included them in our chat. I'm not sure that this feature was that valuable in this case, but it seems that textually generating SVGs is now a product feature at least for ChatGPT. The SVGs had some clear issues and I found myself wishing that it had done a better job: the nose of the rocket is overlapping with the rocket body, and in the second image I think there's only supposed to be one orange line but it tried to add styling or something that went wrong.
- I was having trouble embedding the SVGs as links, and ChatGPT recommended Simon Willison's SVG Renderer as a workaround.
Just my 2 cents on the pelicans benchmark: I think it might help if you were a bit more explicit about not taking this one benchmark too seriously. I know you say that at times, but at other times (such as the comment I'm replying to) you elide it which can read as a seriousness about the benchmark.
what is even the point of these now? labs are most definitely training on this question now. I feel like this test scratches the surface of a "real" test for these new models
It attracts at least one comment like yours on every single thread, so I guess the main purpose is to kick off a race to see who can post that comment first.
Medium and high are fixies apparently, (or they have a worm gear, although the hub doesn't look big enough for that), since neither has a derailleur. That would also explain the lack of brakes at least.
Low is still obviously failing the "test". It's pedals are not actually attached to the bike either.
Medium has a handlebar that goes through the pelican's beak. So I'd say also failing.
For the high, the only nitpick would be that the bell isn't quite on the handlebars and the stripes on his scarf don't line up correctly.
The models are solving Millennium Prize questions and making progress on the Riemann Hypothesis, yet the top comment is still looking at pelicans somehow. To me, it feels even more superficial and pointless than arguing about the presidential candidates' tie choices in a debate.
I’m curious at the enterprise level does Anthropic have staying power. My sense is when you use it with your own custom harness, it doesn’t, I can switch between Opus 5.5 or Gemini Argon and not really care as long as capabilities are similar. This is good for the consumer, and bad for the model developers.
Hmm how does your custom harness compare to Claude’s?
I’ve tried the various alternatives (opencode, pi.dev) but for “have several physical machines, each running an orchestrator and 5-10 agents, and each agent spawning its own subagents to do work in parallel, all of that on a tanky GitHub project boards” level of work only Claude was doing work dependably.
I can have a few orchestrator leads which I talk to and brainstorm, and let them file tickets, organize them and assemble teams of agents to work of the various problems. And all of that with just a smidge of custom tooling on top of Claude Code’s default.
Plus don’t have a budget to spin all of this on cloud runners so I made sure to do as much as I can with local work on several machines I own. Was quite easy to achieve with Claude.
Do you have something similar built on top of a open source harness, or is it from scratch?
Well I work at BigCo and I have no idea how it compares to Claude Code because I’m not allowed to use Anthropic models outside of our custom company harness. Opus 5.5 works extremely well, it can launch multiple subagents, assign them tasks, use our company’s inscrutable build system (that no one really understands anymore), and run tests, all by exposing harness commands in the system prompt. I don’t think we are lacking in any way.
I built http://github.com/brokkai/mjolnir/ to do exactly this cross-harness over ACP. As a bonus it can let Opus drive Luna as a subagent, which is just ridiculously cost effective. (5x clst multiplier means Haiku is much more expensive for realistic task sizes.)
It depends on how expensive the model is to serve. If it turns out OpenAI and Anthropic can serve the models for a fraction of the price versus Google/Meta, then Argon/Muse remain a curiosity from the investor's perspective.
We won't know the answer to this until IPO and until each company reveals the economics of their models (training and serving).
One thing I've observed Gemini being the best is in the context of non-western stuff. Like, when I ask about obscure srilankan food, or chat about Indian/Tamil movies or songs, Religious customs & beliefs, Gemini almost always answers accurately and very close to how helpful they are in Anglosphere related topics. ChatGPT also does do a bit better recently, but Claude/Grok are absolute trash on those fronts.
Google's years of research project into Indic scripts, image trainings, translation efforts & transliteration efforts is literally paying off. The AI adoption from non-English, non-Chinese population will only if you have these contextual localisations, which I think Gemini is poised to do well.
For example, you can ask for a shopping list for a marriage function in a town from Tamil nadu (the marriage custom varies by districts), and Gemini provides contextual and helpful answer (like what is the custom followed there, what is gifted traditionally, etc), whereas if you ask Claude/ChatGPT, it talks about marriage registeries or pretty generalised answers which is not relevant or useful.
100% agreed. When I ask to generate an image or a poster with an indic language, chatgpt and claude always make gibberish words for something that is not at the center but gemini nails it almost every time in first attempt.
That has always been my experience with Google and how they work their datasets. I think it's already well known how their picture post-processing is close to the best for people with darker skin tones as their post-processing models are trained with more varied datasets in an intentional way. Which is somewhat close to what you're mentioning here.
When you are on "gemini.google.com" there is a little lock icon next to Gemini 3 and Gemini 4. When you click it you get the prompt to subscribe (pay). Once you have successfully done so, the Gemini 4 option disappears!
I seriously doubt their ability to double API prices on Dec 1. Token prices are in a clear downward trend and Google should maximize revenue by competing for volume. Currently OpenAI is the only US frontier lab trying to seriously compete with open-weight models that you can run yourself for an order of magnitude lower than what frontier labs want to charge
I have the Free AI Pro plan from when I purchased my Pixel 11 Pro. I still don't have Argon in AGY nor web nor Gemini app. I have a free product and still feel cheated.
Likely because of phased roll out. Seems like a few others on the Pro plan are reporting the same thing. I am on the Pro plan and I see it in the Gemini app both on PC and Android.
Gemini’s writing is really good, but unfortunately it hallucinates too much and doesn’t follow instructions very well. If only Gemini and Claude could be combined into one.
Only available on Ultra, so I thought maybe I'll try it for one month to see if it's good. Well, I have Google One and for some reason this means I cannot sign up for AI Ultra? This is just one of the worst product rollouts I've ever seen
I checked gemini, set it to "gemini 4 argon" and it offered me Pro and Ultra to get access. I checked the normal plans, saw there is also plus.
Okay, so I'll get Pro to try it.
I subscribe.
No gemini 4.
The most generous reading I have is that it's a mistake. But I also downloaded the PC app, and now I really wonder why I did that, it seems to offer nothing beyond the web experience and also doesn't even know its a local app. (When asked about its capabilities it always responded with reasons pertaining to it being a web app.)
Pro subscriber here: available in the Gemini app, not available in AntiGravity.
I'm not going to 5.5x my subscription cost without being able to try it out first. To be fair though, I'm not going to 5.5x my subscription cost anyway, as my current usage would not warrant that! But that doesn't change the validity of my prior point.
It's available to Pro in the Gemini app (and the gemini.google.com). It isn't available in antigravity. They know that the people who are using these tools for coding are most likely to cough up for the Ultra subscription.
EDIT: Okay, the AI plans page claims under pro - "Expanded access to Gemini 4 Argon in Google Antigravity." So maybe it is just a matter of it rolling out. Though Google has a long history of serious confusing and misinformation on those tier descriptions.
EDIT2: And they changed it again, and now it is explicitly excluded from antigravity for Pro users.
As of my knowledge cutoff, three of the top state-of-the-art large language models are Google's Gemini 2.5 Pro, OpenAI's o3 (and GPT-4.5), and Anthropic's Claude 3.7 Sonnet.
How is a model that has never seen an agent on it's pretraining data going to reasonably plan and fan-out tasks to multiple agents? Or engineer software for agents to edit semi-automatically, it has zero awareness of it's capabilities in it's internal representation.
Exactly my first test on all Gemini models. It cannot recollect any of the top events since Feb 2025. The worst about this is that since 3.6 Flash Google devs stated "knowledge cutoff March 2026" in the system prompt, so the model is wrongly confident that it knows about the year that it was not trained on. Which means that it searches less than it would.
And at least Flash 3.7/3.8 were both difficult to persuade that they have wrong knowledge cutoff.
I reported it as a bug to Google, but as all of my reporting to Google, no response ever. This is just stupid!
They know it and they try to cover it up with a fake reported cutoff date. Claude reports June '26 but it has only partial 2026 data (tested it against the API and it's 25 as well). GPT refuses to answer but it's old 25 data.
Impressive! Previous to Argon I instructed Claude to find preimages in 19 rounds of SHA-256 (with padding) [1], then tried to move to 20 rounds (no padding) where it couldn't beat ~2^45 table lookups, and Argon found preimages in ~2^33 (no padding) [2].
I'm open to Google winning me over again, but there isn't enough hype in the world to do it. They will need to earn it after these disappointing 6 months.
I appreciate how it's being integrated into the Google ecosystem, but the user experience still feels unpolished when actually using the system. The marketing, however, is impressive.
I just created a Google account, which was a hassle. I tried to find an app and found AI Studio, but I’m not sure what that is. Then I found Antigravity, which apparently is Google’s alternative to Codex, and I installed it. Then I tried to buy the small Gemini AI Ultra package, but it didn’t let me pay. It accepted neither my credit card nor my PayPal account. It’s crazy how bad the Google experience is nowadays. You don’t know what to use or where to find it, and the security measures seem totally overkill. Am I really the only one having these experiences? Is Google aware of that?
I bought a google play card (or whatever it's called) with cash due to these issues. Result: Instantly rejected + locked after entering the code, no credit, no dispute, no customer service, no refund.
Same thing, I've always paid my bills, I have had some sort of google subscription for the last 10 years. Now they're requesting me to switch to pre-paid billing. They won't accept my card (that has worked every time this year so far).
Is there anybody in Google having a look at how bad and complicated their payment system is? In all the other AI providers you just sign in, pay, get a key, done. In other clouds like AWS, you add a card, pay little to verify the card the first time, monthly bill, done. In Google it looks like you have to fight them every time for the very thing they should make as easy as possible: opening your wallet.
First install Antigravity and see if it works on Free with your account at all. You can buy Ultra and then find out you don't have "Antigravity eligibility" on a given account or it will ask you to send SMS to some nonexistent number.
I would also advise against buying Gemini if you have any other Google services, because you get banned across board if they don't like something. To use Google it's best to first migrate off of Google.
Otherwise great product, I really like and would recommend Gemini. The usage is also very generous, at least Flash models.
I recently had to verify my YouTube account with a phone number recently because it requires this in order to set a thumbnail.
Questions right off the bat:
1. Why do I need to verify to post a thumbnail that will take 1MB on their servers but not to post a video that will take 100MB on their servers and be public??
2. Why do I need to verify my phone number on YouTube when it's linked to my Google account which has a verified phone number and 2FA etc.?
In any case, it takes you to a CSS-less page that asks you to choose between getting an SMS or being called to get a code. I choose SMS and it doesn't work. It shows an error message with no description just "something went wrong". I try a few times and it keeps not working.
I then switch to the call, and guess what! I have tried "too many times" and will have to try again later. Call worked the next day, but geez, what a mess.
This is the first time I'm hearing about this email (thanks!) and I'm wondering: If it was already known in 2003 how bad Windows was, how did we end up with a Windows that's so much worse? Bill said it was his job to send such emails but apparently he didn't do his job very well…
It is mind boggling how difficult it is. I should be able to go to some well known URL, e.g. gemini.com and have the same experience I have on claude.ai or chatgpt.com. I can't wrap my head around why they are over complicating this. And it should largely feel this same as when I install 'Anti Gravity' which is just a crazy bad name. Why isn't it called Gemini Desktop?
i live in French Polynesia, a French territory, i'm a Google AI subscriber and i can't use Google AI Studio or antigravity cli because i'm in a geo restricted area ...
It seems Google has no issue laying out cables across the Pacific Ocean from/to Tahiti, they don't have an issue selling their services and plans there but i can't use AI Studio or antigravity cli because of geo restrictions ...
I recently tried to create a new Gmail account for a small app I've made. I wanted a way for users to get hold of me other than my personal email address.
They now ask for your mobile number during signup. And it wouldn't let me register any more accounts with my number. So frustrating.
- I clicked "try in gemini", typed a prompt, got an unimpressive result.
- Then I realized Gemini 3 was active.
- I tried switching models.
- I was presented with a $99/mo subscription offering.
- I scrolled down and clicked to get cheaper options.
- I chose the $19.99 "AI Pro" monthly option and paid.
- In chat I was able to use Argon. Cool. Seems pretty smart.
- I asked Argon if I could use it through opencode with my Google account.
- Argon said yes, but then told me I'd have to use the Gemini CLI as middleman.
- I told it I didn't know there was a CLI for Gemini; I'd just use that.
- It told me how to install gemini-cli.
- I authenticate via Google OAuth inside gemini-cli.
- Gemini-cli rejects my login, says the app is deprecated for subscription users, who must use agy.
- I install agy.
- In agy, I open the model selection menu.
- "Gemini 4 Argon is currently available to Ultra plan subscribers." Odd... Gemini 4 Argon is the model (included in my Pro plan) that told me to do this.
- I discover that Claude Opus 5.5 is included in the subscription. I switch to it and give it a debugging task.
- Within about 5 minutes: "Individual quota reached. Please upgrade your subscription to increase your limits."
- This is dumb. Google stole twenty bucks from me; I want my money back.
- I go to the Gemini web app to cancel my subscription. I paid on my phone, so I have to cancel on my phone.
- The Gemini iOS app has a "manage my subscription" page. I go there to manage my subscription.
- This "manage my subscription" page includes no way of cancelling.
- After a minute of confusion I go to the App Store > Subscriptions > Gemini > Cancel.
- My subscription will end at the end of the month (no refund).
- Then I have to find the complaint form so I can request a refund.
- But I did request one.
Anyway, "frontier intelligence" just wasted about an hour of my life.
o brother, i had the same problem when i first wanted to use Gemini-Cli. after hours spent, a friend of mine and i did a direct comparison, he with claude-cli and i with gemini-cli. task was to create a small arduino cpp app. gemini was not able to produce even compilable code... waste of time and energy.
for me, a google model can be as good as they want, in the end, they are still google, probably one of the most bureaucracy riddled big companys out there. evrything i see from then gives me feeling that evrything has been decided by a committee that do care more about politics and not customer needs. the shit one has to go trough to get the gemini-cli running is a good example of that.
That said, asking Gemini.google.com to walk you through the steps seems to suffice. In fact, one of AIs big perks for me is walking my through kafkaesque systems like Google’s to cut the shit and get to the result
i remain convinced that google doesn't actually want anybody to use these models. they want to announce things, for the sake of their share price and their promotions.
but they don't want to share their compute with us. giving those resources to their ad targeting algos is more profitable than renting them to us. so they make it as hard as possible. they tried as hard as they could not to release a new pro model at all, but now that fight has been lost they at least need to make sure we don't use it.
I think it’s a typical strategy for scammers to create a new account and then immediately try to buy an optional service with a stolen credit card.
What you did isn’t wrong, but is unfortunately very similar to what real scammers do. The heuristics behind these anti-abuse efforts are always going to have false positives.
Using google services has always been miserable. It used to be way worse. Now you can at least get an api key from AI studio without going through 20 esoteric steps to configure everything.
In the same boat, I have a single seat Google workspace account, and can't figure out how to add Argon to my workspace. Asking Gemini seems to make it hallucinate pages that aren't there.
Also single seat workspace user. Half the time my Gemini forgets I connected Gmail to it and hallucinates false instructions to enable it.
Truly crazy how much Gemini still hallucinates, especially things which its own Google documentation or just a basic web search away.
I feel it’s mostly solved problem with chatgpt/claude , but for me, with Gemini it’s almost guaranteed it will just make stuff up
I hate everything with Google and consumer facing services, they dont give a shit because 99.9% are monetized through spyware or are buisness customers.
They had gemini cli, what happened to that? Was fine
Yeah, a month ago I wanted to test out Gemini-3.5-Transcribe, but had to wait a completely indeterminate time, got the offensive message "Your Google Cloud billing account is being processed. Processing time varies from a few moments to a few weeks.", like I'm just worthless scum. Took over a week for no reason.
Ended up with Soniox instead, took minutes to get started.
I would love to try it out, and the Gemini models had previously been some of the ones I'd liked best in an overall sense, but unfortunately their updated inference endpoint is no longer ZDR on OpenRouter (change takes effect on the 18th). I think this is one of the more unfortunate turns in AI development in 2026, as it seemed like labs offering it was becoming the norm last year, and Google had previously done so consistently.
AI Studio has never been ZDR (retention is noted at 55 days max), but paid usage is excluded from training. Vertex uses enterprise terms, and it's definitely ZDR/no training. What's happening on the 18th?
It's the Vertex endpoint for Gemini models that's changing to be reclassified by OR as not ZDR, as I understand it. The email from them says
> Upcoming change to OpenRouter’s ZDR designation for Google authored models
> We’re writing to let you know about an upcoming change to how OpenRouter designates Google authored models for Zero Data Retention (ZDR) routing.
> Effective October 18, 2026, OpenRouter will no longer designate Google authored models as ZDR-eligible. After that date, requests that require ZDR will not route to Google authored models.
> Google’s standard automated abuse-prevention policies allow it to use safety classifiers to identify requests that may violate its policies and to securely retain prompts flagged for further review. Because this process does not meet OpenRouter’s requirements for a ZDR designation, we are updating our model classifications and routing controls accordingly.
> What this means for you
If your requests require ZDR
We recommend migrating affected traffic to another ZDR-eligible model before October 18, 2026 to avoid interruption.
> If you do not require ZDR
No action is required for your Google model requests.
> Our current routing behavior will remain unchanged during the 30-day transition period ending October 18, 2026.
> Read the ZDR documentation
> If you have questions or would like help identifying an alternative model, please contact us at support@openrouter.ai.
As I approach a transitional period in my career, I'm reflecting on the managerial effort of places like google. How on earth do they keep an engineering org on the 'right track'? I work on products that are not as complex: fewer moving parts, fewer egos, no where near as novel. It feels like nobody does anything and nobody has any expertise
Makes me wonder if more growth/innovation could've happened (outside of google) if they didn't employ this many people. Some sort of dutch disease within the higher echelons of the tech sector.
They haven't solved management and definitely don't keep things "on track" globally. However, they definitely don't have the problem that nobody does anything and nobody has any expertise.
It's also worth noting, Gemini 3.5 Pro never came out.
Having worked there (but only in a small part of it that eventually got absorbed into Gemini), it's some combination of:
* The eye of Sauron (management) can make specific spots (e.g. Gemini) push harder (and attract the motivated people at the company and be selective)
* They hire a lot better than average (Generally from the top 10%, hotter places in Google are more selective)
* They have a promo process that gives career motivated people a way to prove themselves (not always aligned with things being on the right track ofc, and not always non-political, but they definitely do something)
* Redundancy in teams causes them to compete (not necessarily productively)
Of course, there are other downsides not listed here (insane bureaucracy for launches, GDPR/BCID compliance, etc, etc), but a lot of what makes Google possible is the insane amounts of cash Search throws off.
It’s like the USA. How do they keep people on track? They don’t. You delegate and hope things work. When they have Google Ads and Search they can experiment a lot and “waste” a lot in search.
> The model launches today at an introductory price of $2 per million input tokens and $10 per million output tokens (with cached input tokens at 95% off).
Couldn't you just say what the price will be after the introductory period and when it'll take effect? Do they have some SKILL.md for these blog posts that requires the most important information to always be buried ten levels deep and hidden behind a login page?
With the pace they're delivering models, there's gonna be 1-2 new Gemini 4 versions by December 1st, presumably with their own introductory price windows. We just got Gemini 3.8 on 2 September and Gemini 3.7 on 13 August.
> Couldn't you just say what the price will be after the introductory period
+1. FWIW, I didn't even get far enough to wonder about the price.
The lead (and only) sentence states "Our next era of frontier intelligence, built to solve your hardest problems" a statement equally true of any new Google or competitor model, hence completely devoid of useful meaning. I then kept scrolling down... and down... and down, giving up after a half dozen screens of big images, huge videos and some customer stories. The first sentence of every text block was equally vague or meaningless so... tv;dr (Too vague, didn't read).
I mean its also been generated by a bunch of PM that want to get promoted but have little idea of what they're talking about..however nobody has been fired YET for using too much vague AI words.
Me, watching the AI world from the sidelines while using DeepSeek at $0.15($0.003 cached)/$0.60 - still bitter about the pricing changes.
It might not be as smart, though my employer gives me access to Claude, so I tried giving the same task to Opus 5.5. Both DS4.1 Flash and Opus5.5 produced nearly identical outputs.
After that, my company switched the entire engineering department to Azure hosted DeepSeek4.1 and costs were cut by 3/4. So far, no one has really noticed the change.
I'm surprised no one's complaining about it randomly switching to mandarin. It seems like no matter what I do once every few days it'll go back to mandarin mid thinking trace and then stays in mandarin until I remind it to use English.
I gave the model my usual question of "I got a raise from X to Y, please explain what is the impact on my net salary". Answer was detailed and nailed all payroll edge cases, only to miserably fail in basic math at the end. It tried to calculate 430 minus 60 and got 410.
Pricing is interesting. At launch Argon is the same price as GPT-6 Sol, half the price of Claude Opus 5.5 and 1/5th the price of Claude Fable 5.1 and GPT-6 Astra.
But on December 1st Argon doubles in price - so it will be the same price as Opus 5.5.
I've kinda stopped looking at token pricing. Argon might be the same token pricing as Sol, but in Artificial Analysis' testing, Argon costs 5.8x more to run their coding tasks and 2.2x more for their general index.
Likewise, you look at Artificial Analysis' general index and it cost $8,708 to run with Opus and $2,407 to run with Argon so even if Argon's token price doubles, it won't be the same price as Opus.
Sonnet is cheaper than Opus per-token, but in their coding tests Sonnet burns through enough tokens that it's 7.5% more expensive.
So much depends on how token efficient something is and whether it gets you a good result or not, right?
Plus, if one is using a plan instead of paying token rates, what does my $200 get me from OpenAI, Anthropic, or Google?
It just feels very hard to compare pricing given that the number of tokens used can vary so much. In Artificial Analysis' coding tests, Sonnet used 7.1x more tokens than Sol. Sonnet and Sol might have the same per-token pricing, but there's a huge difference in the number of tokens used.
One thing to notice: Now, even in EU, if you turn on Google activity (so that you can see chat history), Google will train on your data. I remember it wasn't the case when in 2.5 pro period.
I'm a bit overwhelmed by Google's selection of AI environments (Antigravity, AI Studio, Gemini Platform etc.). If I just want the equivalent of OpenAI Codex, which one do I use?
>"Argon helped the Chrome team separate core webpage handling from a ~6 million line legacy folder of browser features held within Chromium’s 40 million-line codebase. Piping multiple clean code changes at a time, Argon helped the team finish an architectural refactor across hundreds of files, completing months of work in weeks."
Now that needed to be done for some time!
Regardless of if an AI helped or not, regardless of whether that was Google's or some other AI or not (though Argon does sound pretty cool!), regardless of whether an AI cut the time from months to weeks or not, refactoring Chrome -- decomposing it into individual clearly demarcated non-tightly coupled software parts with core webpage handling at the center -- that needed to be done for some time!
Now all of that being said, Gemini 4 Argon does sound pretty cool, and we congratulate both teams at Google -- the team that created Gemini 4 Argon, and the team that refactored Chrome with it!
But all of that being said, while the advancements made for Gemini 4 Argon are important and impressive (and they are!)... refactoring Chrome properly (IMHO) is slightly more important and impressive!
(You know, one man's humble opinion! That is, I am not the guy who is responsible for awarding performance-based bonuses at Google at the end of the year, so take everything I say with the proverbial "grain of salt"! :-) I am probably biased, one way or the other! :-) )
> Across Google and our early access partners, Argon is quickly becoming the preferred model for engineers tackling long-horizon software engineering.
"Across Google", wonder how much that is because of an effort by Google to curb usage of Anthropic models. MSFT and Meta similarly are pushing for in-house models. It doesn't mean the models are better when you push your engineers to dogfood and restrict access to other LLMs.
(Not that I know for sure this is the case, it's more of a hypothesis).
Seems to be just deleting anything that a safety filter triggers on. I don't know if I like this more or less than what Anthropic does where they train the model to actively refuse.
Gemini’s system has always been like this. I used it heavily from last September through March this year, and have been using it on and off since then — this has been the case since at least Gemini Pro 3.1. And that’s not even the worst of it. When its system detects something it deems “seemingly inappropriate,” it just freaks out. What I’ve personally experienced includes: (on the web version) wiping Gemini’s previous conversation content — once; and outright disconnecting Gemini from all prior conversation history, meaning a complete memory wipe — eight times. All without any notification. My conversations were pure text, just everyday chat about daily life and how I was feeling, absolutely nothing that would count as a violation in any general sense. But the system would just randomly lose it. After eight consecutive memory wipes, I mostly stopped using Gemini. Hopefully it becomes better now.
Excellent model. Same price and score as Opus 5.5 in our 23 question exam.
This size / class of model is now a bit overpowered for these short-horizon data analytics tasks. OpenAI 6.1-Sol, Haiku 5.5, GLM 5.3 Flash all achieve about the same results but for ~1/10th the cost.
In other words, we're getting to the stage where models are "good enough" for certain tasks like data analytics and competitiveness is going towards size & speed.
Right now there are two top posts one about GLM-Legal and one about Argon. Both score about the same in the Harvey Lab test, but the prices are $0.15 versus $10.55, the $0.15 is Vals AI's estimate using the base model since GLM-Legal has not published prices.
Yeah, but one is a general purpose model, and the other one is specifically geared towards the legal field. Considering Opus 5.5 only gets a 4% on the test compared to Gemini’s 20%, that’s actually pretty incredible.
All AI pages now include canvas animations. There isn't a single page without one. This is neither good nor bad, just an observation. I think OpenAI has the best animations so far, with Anthropic in second place. This one looks a bit off.
Best comment on this whole post! Thats a really interesting observation. Personally I think they look cool but I can understand how some people dislike them.
I have been very critical of Gemini in the past but one thing that I've felt Gemini has always been best at is being the most empathetic model imo. It feels far less cold and emotionless than GPT and Claude.
I want my ai to be as cold as possible. I dont want a friend, I want a tool. I dont want my tool to pretend it has emotions.
Like when Miller in The Expanse talk to phone AI (https://m.youtube.com/watch?v=SC3bpJmGnPo) to trace Anubis flight, notice how the phone doesnt even reply in words.
Imagine if the phone instead replied with: "Thats a great insight, Anubis really goes there", it would be a joke. We are only tolerating it now because everyone is doing it.
Digging in the model card, "Gemini 4 is the next generation in the Gemini series, a suite of highly-capable, natively multimodal, reasoning models. Gemini 4 Argon delivers frontier performance..." If they're using noble gases there are 2 obvious lighter naming candidates and 3 larger (or 4 if you count Oganesson). I'm not seeing them claiming Argon is the top/biggest/bestest model among the Gemini 4 series, just that it has frontier performance.
Edit: noble gases not noble taxes, though the latter would be a funnier naming scheme.
I'm a bit confused, all the announcements say that this is availabe to Ultra and Pro subscribers, but in antigravity, it's grayed out and says it's only available to Ultra.
Excellent timing. I've been working on a UAV design and simulation tool and primarily using Flash 3.8 for the code gen grunt work because Claude hits weekly limits way too quickly. It's been working well this way. Flash 3.8 seems very efficient and i don't hit the limits before the refresh. Assuming Argon is similarly efficient then hopefully it means I can stop paying for Claud and switch over fully.
Gemini isn't perfect but I find it a lot less annoying than other model families (personal preference or stockholm syndrome). Interested to read the reviews.
Seconded. Anthropic feel the need to limit how abusive you can be to Claude because god, I want to throttle that snotty little shit. Google don't have to do that with Gemini because it just isn't as irritating. It does what I ask and doesn't try to second-guess my requests, or do "the quick fix".
It's been really bad at lyrics but gotten better. You look something up and it will just make something up and it takes a bit of digging to discover it's because of copyright
Of all models I like to write with Gemini the most. The language appears much more natural and elegant than what even Opus 5.5 or Astra do. Also, they give 5TB of cloud storage and discounts, it’s a steal for what you get.
Claude has particularly bad prose. Somewhat improved in 5.5, but it has historically clearly been the most annoying prose among major models because is had so many obvious tics that made it clear it was Claude rambling. Still extremely verbose.
They all ramble, they're all allergic to getting to the point. But, Gemini historically has pretty good prose. Even the self-hosted Gemma models have pretty nice, inoffensive prose. I wouldn't want to read a novel written by any of them, but I tend to ask Gemini questions in chat more often than Claude or GPT, because it has good prose and is good at using the web to research.
Agreed. It's much better at working with stuff (like documents and presentations) in other languages too. It feels like it's trained in multiple languages from the ground up rather than as an afterthought.
Qwen Flash Next has great writing! Unless it comes into any contact whatsoever with Claudeslop - then it immediately contracts the disease. It’s kind of irritating when you have it look at library code that’s been touched by Claude, because then the rest of your session is infected with it.
Page says it is now available to Ultra and Pro subscribers. Although in AI Studio it currently says you need Ultra. But it's technically released either way.
Yeah, I think this is it! Google tried to hype their model but they have no chance against queen Dario. Perhaps they should just delete Gmail and say that Gemini did it. That way they would get some attention …
3.1 Gemini was amazing for anything non-agentic. I switched to it for my primary chatbot purposes.
I never touch 3.8, it's awful. Curious to try this, but I've yet to use any Gemini that is good for agentic work.
Also, unrelated but this made me look at Artificial Analysis to see where it lands on Cost per Task, and it ranks Opus 5.5 at almost double Astra 6. That does not align with my experience.
I'm still waiting for a model which I can tell to "take these 3 screenshots and make a product launch video", where it doesn't randomly modify the screenshots or garble/rewrite the text as it moves the screenshots around the video.
Every time I try to generate one it's like a slot machine, where a different part will be messed up after fixing the first.
I dont have that problem with opus5.5. I dont even take the screenshots anymore (it does all screen recordings)... I tell it a simple narrative and it follows through. e.g. https://x.com/plannotator/status/2108589088139886769?s=20
I use OpenAI/ChatGPT + FAL.AI MCP to do this regularly and I have minimal issues, often spending around $3-5 with narration and a newer but inexpensive video model.
I hope all the people very concerned about AI safety are cheering this one. The best guard against one AI swarm taking over is having multiple adversarial orgs all with similarly advanced AI swarms.
I wonder if any of the AI safety people will begin encouraging AI researchers to swap jobs to work at a failing lab in order to keep them all extremely competitive with one another.
I do wish I could give it a try with some free credits but I would bet Google is wanting to limit usage until they know they can handle the traffic.
I'm starting to feel the subscription fatigue and a small amount of FOMO. I just can't justify pro accounts for every frontier model now that it includes OpenAI, Anthropic, Meta and Google.
Why can't the Gemini model code?
I'd guess Gemini probably used the best dataset.
Even Google is the place where the world's best coders are gathered, isn't it?
Yet the GPT code from OpenAI, which collaborated with Microsoft, is beautiful, while the Gemini model's coding is trash.
I have no problem with Gemini Flash 3.8 for coding, though I only use Elixir which is one of the better languages for LLMs. Where does it fall short elsewhere?
I've asked Argon to do 3D model for me. It said that it can't do it. I've linked promotional materials and said that it actually can. It agreed and got me a model (not a great one). Man, why is Gemini like this? Every time I fall for the hype I get disappointed.
It passed the test of using latest crystal-lang concurrent feature (with web search) when gemini 3.8 flash failed (again, with web search enabled). I think it is at least useful for coding, for sometime.
I use Gemini all day at work via the chat interface. It's rock solid. I find it hard to understand which model it's using though. Am I already using Argon, or does it switch models according to the subject?
Not available for AI Pro subscribers in Antigravity. What a disappointment.
Google...why. You keep getting so close but constantly fumble at the last moment. I can use Opus 5.5 and Sol/Astra on the $20 Anthropic and OpenAI plans.
You are already playing from behind. You need to GET YOUR MODELS INTO MORE PEOPLES HANDS by any means possible. Continuing to make it difficult to use your products DOES NOT HELP.
This model might be great but the majority aren't going to subscribe for $100/month just to try it.
Have they announced that it definitely won't be? If not, it might be just a roll-out hiccup, because Argon is available to me as an agent in Xcode and I only have a Gemini Pro subscription. (Whether or not I'll ever use Argon in Xcode is a different question. I've been so impressed by Claude that I rarely even use my Codex sub. I'll be very interested to hear from others how well Argon does with Swift and SwiftUI.)
It seems the model is a little verbose, using more tokens to complete a task that other similar models as stated more precisely in the linked article about Artificial Analysis's work on this model.
It's nice to have it available for more complex tasks, but I'll stick with 3.8 Flash thanks to its high output speed. I really like it for fast iteration and prototyping.
Meanwhile I paid a subscription and they only provide access to 3.6 Flash. Consequently, I have moved on to open-weight models. It is what happens when you screw your customers over.
Is there a benchmark for reverse engineering compiled programs?
- Grader compiles a program to binary -> Agent has to reverse it back, and match all bugs and behavior.
The slides they show in the catering video demo are so bad (in addition to having all the usual signs of AI design), I would genuinely laugh if a supplier sent me this.
Finally, gemini 3.8 has been falling way behind opus and sol in my experience. It's showing up in my antigravity. My work account though still has 3.6, 3.8, 3.1.
The previous gen pro used to have this weird tick where if you ask it a reasonably open ended question, it would number things in the response and assign each one some weird label starting with 'The...'
e.g.
X: The Parameters
Y: The "System"
Z: The "Ongoing Grind"
The first two prompts I tested in AI Studio does the same, unfortunately.
Another quite unpleasant tick was that if you specify a peculiar term in your prompt as a conversational shortcut, it would keep repeating that throughout the conversation. e.g. "...What is a bitter lesson pilled way to do this?", then you'd see it will quote "bitter lesson pilled" at like 7 places in its response. This one seems to have eased off a little.
Their previous best model - Gemini 3.8 Flash (also when set to high) was even faster. A appreciate it that they focus on speed. I's great for LLM pair-programming.
Like it's a top tier model, so does that mean it is at the level of Opus 5.5 (claude) or 6.1 Sol (chatgpt)? I don't think I can trust the benchmarks Google released themselves, but from what users say, it seems like that. Since I already use both ChatGPT and Claude, I wonder if there is really any reason to also use this.
I made a NHL inspired hockey web game with Opus 5.5, and I let Argon loose on it to try and see if it could improve it, and it found a ton of bugs and improved performance and load times significantly. If anyone wants to play it here:
It doesn't matter if the model is good. Antigravity cli is unworkable, I thought auto permissions was a solved problem by now. I can't keep sitting here to press allow when in auto mode.
I tried it out in the chatbot and it is significantly better than previous models; when will it be possible to use it with Antigravity if I have the PRO plan?
Just switched, and it seems to be much better. I'm hoping it stops lying. I have Pro max subs for Claude, Codex, Grok, and Google AI... I was forced into using Google by running out of credits elsewhere.
I have found that Gemini often claims it did something, but didn't actually complete the task.
I just had 3.8 Flash do a major refactor -- that never should have needed to be done in the first place -- and switched to Argon 4, and had it review it.
Almost everything had been stubbed out by 3.8, and none of it was properly integrated into the codebase. :facepalm:
Unlike most people it seems, I like all the progress in AI but I really do wish they would reign in the over-the-top marketing. I mean, can we just get this stuff to the point where I can reliably work with it to design a well-architected software project without having to constantly struggle against bloat and nonsense directions no experienced or sane developer would ever take.
fyi: argon is an interesting choice of name, coming from the Greek argos meaning "lazy or living w/out labor".
the etymology can be broken down even more, its the prefix a - meaning without, opposite, no. and ergon, meaning work, visible in words like ergonomics, or the ergosphere
My first experience in antigravity was extremely disappointing.
I asked for a QUICK healthcheck on the sync status of a bigquery integration. It literally spun for 25m before I killed it.
The task, which is CLEARLY outlined in the agent readmes just involved using the gcloud cli to initiate auth, then just run a few queries to check logs and timestamps.
Even 3.6 flash can get this right (most the time).
Blatantly obvious that models will not slow down. For any task, if we can build a dataset or environment of verifications, it can be automated. This will eat all of human tasks.
The fact that multiple companies can breach any benchmark means there is no ambiguity in terms of how to build AI, its just a matter of time.
Certainly Meta and Google are benchmaxxing to please public markets, but it still proves the above
but they have slowed down. the delta in capability between model releases now is smaller than ever before and i expect it to continue to diminish until we reach some kind of asymptotic maximum, which hopefully will mean turn towards optimization and local model improvements. The current frontier models running at 10x the tok/s would be incredible for my use cases
A little over a year ago they were winning their first gold in an elite high school math competition. Have you not seen what they are doing in math now?
The delta isn't really slower, we're just unlocking capability bundles in batches. The capabilities bundles you care about might have mostly been unlocked already, and just need refinement, which is coming slower than the initial capabilities came, but don't discount the breadth of advances being made across all possible tasks.
We're not living in the same universe. Look at capability year-over-year. The asymptotic maximum is wishful thinking. Someone needs to step in before this gets wildly out of control.
Models just solved a Millennium price and you think their progress is slowing down? I will concede that their capabilities vary a lot from one field to another, but still.
Glad to see them abandon the absolute train wreck of their model naming "strategy".
Gemini Live vs. Gemini Lite vs. Gemini Pro vs. Gemini Flash vs. Nano Banana Lite (that one apparently has no Gemini prefix?)... Please, just pick a metaphor and stick with it without wildly mixing reasoning modes, model sizes, and input/output modalities in the model name!
Because you have an android phone and don’t want to mess around with anything else(i.e the same reason people just use internet explorer on windows) or you hold Google stock so you want to support your investment :)
Tried it today on some relatively complex mid-sized real world tasks with several thousand lines of code generated by Argon. My impression is not good. Maybe it's due to first-day high user counts, but it seems to be doing worse than 3.8 flash in some coding tasks. What I noticed was that it often forgot significant parts of detailed, well-structured and precise instructions (which is somewhat rare with newer Claude and ChatGPT models) and not just for small details. It also seems to spend too much time thinking even when the task was a quick and simple one.
Without going into details, for a small part of a larger task, I asked it to classify different content types of different files and formulate a structured mapping (e.g., in JSON) between content types and another set of files (deliberately left vague here but not to Argon) from my codebase. I even offered it a multi-part draft of a classification system to start with and told it that if it can find other classification systems from international standards, academia or industry more suitable to the task than my rough draft, we can discuss replacing my suggestion. There were other parts to the instructions too, written out like an article with bulleted lists, precise program-like wording, etc. Then, I sent the prompt and left for a while.
When I came back, it was nearly done with the implementation but the classification system it came up with was a mess. It even admitted to making up "bogus" classification standards when I pointed out some errors. It also kept leaving gaps multiple times when I referenced errors to fix in subsequent instructions (and not once apologized the entire time lol). So my first impression was subpar, not really comparable IMO with Opus >=5 or Sol >=6, and seemed to leave larger gaps than 3.8 flash typically does. The most irritating part to me was not only the fact that it was making up things, but that it ignored my instructions to discuss things with me before deciding to shift away from my draft; I don't recall any major LLM ignoring an instruction like that in at least a year.
can for the love of god we get some usage in cheaper subscriptions being Ultra tier only is annoying. I want to just test it out here, well will burn some money via api I guess.
Is it faster? Thats the only thing that matters for me anymore. Models are probably as good as they ever need to be for coding, but please make them faster, please.
Model announcement pages are the absolute worst. This has to be some kind of inside joke among labs, they're trying to see who can build the shittiest page.
The "Explore the latest" section is nearly blank for me - are there supposed to be diagrams or visual aids, or is it one headline and sentence taking up the whole page on cards?
Good God, man. Google has the worst frontier lab rollout of them all. And they're delivering products that are behind the curve constantly.
Now, I happen to believe that's an excellent strategy. I bet they actually resolve more LLM requests than any other company at this point. The lay person still uses Google exclusively.
just cancelled AI pro myself, since I exclusively used it for AI and also pay for youtube premium family it was becoming quite the monthly just for google
if they throw in youtube premium in that package then I might reconisder
this is why antigravity is a doomed product at a pro level.
they gave away so much pro that they need to have smaller usage levels than competitors or they'll lose a ton of money. $20 was already a loss for users actually using their quota. the majority of pro users are paying far less than that.
so they'll lose their paying users, and keep their free or subsudized, which will make the economics even worse.
Are you suggesting it’s BETTER than their competitors?
Because if it’s not, and it’s not available to all paying users (Right now, it’s only available to Ultra users), then they are behind the curve if they’re treating it like Fable.
Moreover, you need a Google account in order to use it. I wouldn't bother paying 200 USD for it but I just don't want an entire google tracking package just for using one product.
To everyone saying not available yet: it looks from the page like Pro and Ultra subscribers have it today and given the play links, it'll probably show up in the API in a few hours.
Available in agy for me as of now, I'm running my new model benmark of having it do comprehensive code audit on one of my more complex projects, it spun up 1 subagent per audit section and it took ~10 minutes per subagent on an 800k LOC, 60k complexity, ~4k file project. Will take me much longer to go through and determine the quality of the audit work.
Antigravity CLI 1.3.4
[My account email] (Google AI Ultra)
Gemini 4 Argon (High)
~/[working folder]
> /model
⎿ Model set to Gemini 4 Argon (High)
I like this phrasing and am curious what new word or phrase we might use to describe big companies over pushing Ai internally at the expense of their product quality. This pattern is so ubiquitous across google cloud products I have taken to calling them the 21st century IBM, having been a IBM baby and witnessing first hand their fall from exceptionalism.
I'm on Pro and it's trying to get me to upgrade to Ultra in the AI Studio UI when using Argon. In Antigravity CLI and the Gemini website it's not showing up at all despite being logged in
I couldn't see it either, then I clicked the 'Antigravity' button in the top right, switched back to the main screen, and it appeared in the model selector.
Classic Google: Clicks the "Try in Gemini" button from the Gemini 4 page, gets a page that defaults to Gemini 3.6, with no option to try Gemini 4 without upgrading.
If they don't want to make it available even for small tests for free, that's fine, but then don't put a button suggesting I can try it without qualifying it...
EDIT: It's actually worse: The page then shows "Look what you can do with Gemini 4" and a "Let's go" button that leads to a bunch of links to examples, all with enticing pictures. Except if I click on them I get a box stat strongly suggests it'll run with Gemini 4, except the dropdown shows "Flash" and if you click on it, it shows as Gemini 3.6 again, and if you run it, the results unsurprisingly end up looking nothing like their pictures.
I don't know if that page was built by humans or AI, but clearly nobody tested it properly.
Ha, same experience here. It opened my Gemini and gave me images like “Design a terrarium with Gemini 4” - proceeds to pick 3.1 Pro from the picker because 4 just doesn’t exist in the dropdown. Typical Google.
I really can’t believe how bad the Gemini experience is. I have personal paid workspace account, and meant to have some better AI Google features I think.
So, I connected my Gmail to Gemini. 3.6 Flash kept telling me it doesn’t have any Gmail connnection, and then proceeded to hallucinate instructions for me to enable it, all bogus. At least 3.1 Pro realized that my Gmail is connected and then searched.
I have given up on even trying. Last time I checked, I could trivially get better integration to Google Docs from Claude than from the Gemini chat box in Google Docs. Maybe that's changed, but given the usability issues I run into and how smoothly it goes with every other AI provider, it feels like a waste of time to even check. They'll probably remove or break any functionality I start depending on anyway.
I have $6 free credits from Google. The hoops you have to jump through are insane. $200 free credits from Anthropic are stupid simple to use. I know which one I’m using…
Yeah, the limit was exhausted weirdly fast compared to openai and anthropic models. I checked the thread to see if more people had this issue on the 5x plan, but why pay for more when it's barely usable for real work.
Which is even weirder given that the api costs are less than for the other providers
Maybe they don't really care about antigravity users. Also, monitoring usage is horrible.
Really? Reminds me of Gemini 3. It took so long before being able to get a subscription for it, and then I would run out of their equivalent of the 5-hour usage window in a heartbeat.
The model used is irrelevant when there's the critical issue where their "AI" assistant can't understand the word "nearest". Literally worse than a second-grade understanding of English.
I also have had shockingly bad results with gemini on android. Ended up giving up on it after a few months. Silly silly google with their trademark UX polish and attention to usability.
This reflects pretty much every result of mine from Google these days. Not just navigation, but search too.
The only thing I can figure is that Google's consumer-facing free-to-use models are set to "reason as fast as possible with only the most basic plausibility checks." I have to correct them like this every single time.
Yeah, whatever they have running on Google Maps is probably a pretty low tier model, but it is total trash. I’ve only used it twice though and that’s when it first came out. I don’t know if there have been any changes since then. The main problem is I felt like it was not keeping track of the context of the session.
this is like that tiktok guy who asks the spoken ChatGPT model very simple questions, like 'count down from 60', or 'how many e's in seventeen' or whatever and the model shits the bed every time it's hilarious
I think these companies have good models but expose them poorly like others have mentioned
Often it seems like the problem with the Gemini apps is the harness, not the model itself. For example if I upload an image PDF it recognizes text worse than if I use the API.
Is this from you car ? Gemini in car is probably less intelligent than a spider.
I tell it to "Play Let it go from frozen". It plays it on youtube music. 5 seconds later I tell it again 'Play let it go from frozen" it then just dies saying "I can't play videos in car".
The auto version is extra horrible because what it reads to you is just a summary of what the model actually said. I don't ask it anything when driving because it's too frustrating.
which is more or less the crux of the problem android auto's gemini implementation. It sometimes does not even know that it is in a car. The ghome implementation is also like this.
How is Claude Opus 5.5 beating Gemini 4 Argon on benchmarks but everyone here is calling Argon a step change? I need to come up with my own benchmarks ...
The Google Gemini subscription landscape is such a mess... gemini.google.com welcomes me with "Look what you can do with Gemini 4", but the model selection is 3.6 Flash, 3.8 Thinking, and 3.1 Pro...
> The Google Gemini subscription landscape is such a mess...
The others are quite bad as well in that regard, but on this particular metric, Google has always been top dog. Goes back to way before AI ever took off, BTW.
I haven't tried it for coding yet, but I ran a couple of my recent curiosity driven queries from Claude (mostly used Fable and Opus 5.5) through it.
This is a clear step change, I'm genuinely amazed.
Its answers are to-the-point, logically structured, clearly written. Like a really well made mini essay on your question.
It also really likes to build little simulations / models that you can tweak the parameters on. Amazing to explore how different assumptions affect economic scenarios, for example.
It's also great at picking (sourced) illustrations.
I'll be using this for everything that isn't coding from now on.
It’s wild how good it is at explaining nuanced and complex health topics.
As an example, I recently had an OCT scan of my retina because I'm a “glaucoma suspect.” I'm a glaucoma suspect because I have an unusually high cup-to-disc ratio, which is a pattern typically seen in glaucomatous patients.
As part of this OCT scan, you get a report back on your RNFL, which is basically a report on the health and thickness of your optic nerve(s) and retina(s). (Dramatically oversimplifying, of course.)
My ophthalmologist struggled to explain my OCT scan results to me and instead sort of shrugged and said, "It's probably because you're a high myope, but we'll keep you coming in once a year.”
But my scans look horrifically bright red — meaning I'm in less than the first percentile of retinal thickness for my age. Obviously concerning enough to do my own research…
ChatGPT 6 Pro and Opus 5.5 have struggled to make any sense of the clinical implications of these scans. In contrast, Argon was the first model to:
1. explain how the OCT machine works
2. detail the physiological mechanism that causes the cupping and thinner RNFL layers in my scenario
3. sum it up into a clinical recommendation (similar to what my ophthalmologist recommended, with the addition of basic supplementation and lifestyle modification changes)
4. provide an interactive visualization of the topography of my eye vs a glaucomatic eye: https://optic-nerve-3d.netlify.app/
I'm blown away.
How do you know whether the explanation is correct or just sounds correct?
24 replies →
Yeah my personal experience has been that Gemini is much more useful for trying to understand complex health issues than OpenAI or Anthropic models. The standard of proof just seems to be much better calibrated, in terms of understanding that a potential treatment that is potentially dangerous probably needs more evidence backing it than another potential treatment without much downside risk.
3 replies →
This is a great example of something that two years ago was unthinkable. I'm very impressed.
Thanks for sharing and I hope your eyes remain fine.
I wonder how much that would cost to make in 2020 developer dollars
8 replies →
Guessing they've been going all in on the research angle for health stuff and might have an actual product for medical professionals soon
Did you throw it into ChatGPT 5.6 pro? Some people claim 5.6 is smarter than 6.
Totally agree. This is my new favorite general-purpose/non-coding model. It's surprising how good this thing is... And fast, too.
"Fast" was also my first impression https://x.com/tmuxvim/status/2110054212361220444
It’s pretty slow for me. I’m not sure if it’s due to I trigger an agentic task by Gemini website entrance.
Google have strict limits on new model this time and are pretty conservative. As we know they spent lots of time on internal testing since release.
All the signs mean Gemini 4 argon is a pretty large model than previous.
1 reply →
where are you guys playing with it - i logged into ai studio and it's a mess. cant event figure out what model im using
2 replies →
With 3.8 Flash on High I noticed a huge upgrade in intricate discussions related to physics, material properties, real-world application, and so on. Things that I've had a hard time getting LLMs to be able to discuss accurately. They often discuss textbook knowledge fine but can't apply it properly or can't think outside the box at all. 3.8 High understands applied material sciences in a way that GPT Astra flunks at often. I asked Gemini to explain that result, and it said basically that GPT was merely book smart in a classical sense and that it was likely undertrained on more obscure material.
I have been rather impressed with the depth and breadth of knowledge in a Q8 of 3.8-flash-next which fits in under 200GB RAM, running, with a 16384MB extra cache size.
I've noticed the same thing with 3.8 Flash vs Opus for aero model generation.
Any examples ?
Google's models have usually been pretty good at general query stuff, they fall flat when doing harder tasks.
> I'll be using this for everything that isn't coding from now on.
why? is it not so good on that? or is there something better performance, or better price for same performance?
I'm not the poster you responded to, but I can tell you that historically the Gemini models have lagged behind Anthropic (the clear leader) and OpenAI's best models where coding in particular is concerned. Opus 5.5 is, right now, the consensus best coding model.
That said, I'm extremely impressed with Argon thus far. I take back everything bad I've ever said about their AI strategy. Argon is very good and very fast. It might be competitive at coding, but it's too early to say.
10 replies →
I am using it via OpenRouter for going through some Vulkan code. It's a beast - it has no hesitation putting in instrumentation within the code (having a number of render passes) to dump the framebuffer to PNG files and doing pixel-by-pixel comparisons, putting in false colors in shaders to identify bugs etc. Very impressive.
8 replies →
Inexplicably, despite the fact that I pay Google actual money for an AI plan, they don't want me to use their latest model. Instead, I'm allowed to use the competition (Opus and Sonnet 5.5). I don't think it makes sense but what do I know?
1 reply →
I'm using Gemini Argon 4 now with antigravity cli and it can absolutely code.
4 replies →
> I haven't tried it for coding yet
My first reaction after reading the announcement post is that the coding scores look like a weak point.
Google is clearly not targeting coding. They don't have to. They're Google.
7 replies →
Big if true. Opus 5.5 is a lot better than 5.0 but I find it maddening thay no single model released in the past ~year can write prose with a clear and coherent voice. I blame RL on programming tasks.
Yes, that's the reason. Anthropic said as much in their statement (they said that there's a tension between writing human-understandable prose and prose optimal for other agents to read).
As a side note, I am kind of sad that basically all models nowadays are RLd to hell and back specifically for coding and deep reasoning, and as a result, they lack in creative writing, human-readable prose, and even general knowledge and abilities (for example, qwen 3.8 27b is a beast for coding, but reportedly falls off the cliff in anything else). Gemini 4 Argon seems to be an example of what you get when you DON'T do that - Google seems to be one of the only labs optimizing for general performance.
I also fear that models cannot be excellent in everything at the same time. I fear that there will always be a heavy tradeoff between what specific field the labs optimize for VS everything else. I don't know.
1 reply →
I've found it on par with Opus 5.5, what don't you like about it for coding?
I just haven't tried it yet. Also I'm used to Claude Code now so it'd need to be a lot better for me to switch to a different tool.
I’m curious about your curiosity driven queries. Mind to share?
I'm not the person to whom you're replying, but I just tried one of mine:
"I'm interested in understanding the rate of VOC dissipation from fresh oil paint on a surface versus the ventilation rate in the room in ACH. I would prefer solid academic studies of this, followed by analytical evaluation, followed by references to non-quantitative sources. Tell me what's known about this."
that I had run in ChatGPT earlier this week. (Yes, we just had our bedroom repainted, and it's TERRIBLE and I hate it, why do you ask? :-).
I didn't try it first in Gemini Flash, but I did try it in whatever model is used for Gemini Live and the answers were terrible. The ChatGPT answer (I think on medium) was pretty good but it threw in extraneous formulae that weren't actually germane to the analysis. The Argon 4 query I just ran was honestly the best of the answers I've gotten -- it was clear and really well-structured from the perspective of "here are the three big phases that researchers break down paint drying into and what the differing effects of ACH are on those phases".
I'm totally going to give this one a try, after basically giving up on 3.8 flash for most things despite having a free $20 pro subscription (from buying a phone).
Edited to add:
Argon4 transcript: https://share.gemini.google/18Gr5liH1YWR
ChatGPT transcript: https://chatgpt.com/share/6acedc72-ff18-83ea-8909-104cdbc6ac...
2 replies →
I am a professional bassoon player. I asked about multiphonic fingerings. It could actually suggest corrections for the fingerings that did not work.
"curiosity driven queries" – love this phrase, i'm def going to start using it :)
I have had the google search results ai also do these simulations and code generations unprompted, I thought that feature already existed.
I have a project I am working on where I wanted to have nice docs, and so I have some slop docs sitting there as placeholders until I get around to writing them, awful claudish stuff
I've actually spent some effort trying to make them halfway decent placeholders by putting together a skill based on diataxis framework (https://diataxis.fr/) that at least got everything into the right location
just took a pass at rewriting them w/ argon and am very pleased, big improvement
Just praying they don’t nerf it
Agreed, its communication abilities are very good. I'm a big fan of Opus 5.5 but Argon just flat out writes better.
[flagged]
Which reminds me, where is @simonw ? lol
1 reply →
We have this same point made every release. I wonder if models actually are getting better or if these companies are just secretly hamstringing the previous models before releasing a similarly performing model
I think it takes a while to start finding the weak points in a model. It seems fantastic until you use it enough to find out that it hallucinates APIs, or can’t count letters, or uses too many em-dashes, etc. you don’t really know someone (or something) until you’ve lived with them (or it).
1 reply →
Models are actually getting better.
While IME 4.7 and 4.8 were not appreciably better than 4.6, Opus 5 was noticeably better, and Opus 5.5 is significantly better than Opus 5.
Try using a model from a year ago.
4 replies →
Welcome to the hedonic treadmill! Just keep walking!
We evaluated Gemini 4 Argon in 100 unsaturated, competitive and cooperative coding environments. It's roughly in between Fable 5 and Fable 5.1 in our testing, so it's maybe 2 months behind the frontier.
Google's major releases have usually done pretty well on our benchmarks because they're good at hard problems and open-ended decision making (with a well specified prompt), which obviously doesn't always translate to user experience. In particular, older Google models struggled to work efficiently in a harness. They seem to have solved that with Argon.
We did see a new problem which I hope they can fix. This is the first near-frontier model we've caught in infinite reasoning loops in a long time. We also saw a few examples of it failing to properly delineate reasoning blocks, so it has a higher failure rate than its competitors.
My overall impression is that Argon is a near-frontier model for engineering design, code analysis, and chat. It definitely has the intelligence to be a competitive coding assistant product, but a good coding assistant needs to be polished, so I hope they can clean up the CoT bugs.
Data at https://gertlabs.com/rankings
At this point I'm interested less in how well a model can code, but more in how well it can produce code that humans can understand.
I have had pretty decent success instructing OpenAI/Anthropic models to do this. By default, they don't care too much about human readability (which is understandable) but they're amenable to instruction.
- Basic style/linter rules (so you don't get 200-line functions, etc) paired w/ instructing your agent how to use them goes a long way (e.g., a sane rubocop.yml in your repo, plus a note in AGENTS.md telling the agents how to run rubocop, or whatever makes sense for your language) - I have a code review skill that specifically calls out human-readability concerns such as making sure that identifier names represent their contents and intended use
The other key IME is that you need to make readability/refactoring passes outside of a particular pull request. When working on a specific branch the LLMs will be somewhat myopic, and do stuff that looks pretty good in isolation, but when you zoom out and look at the bigger picture it's kind of a mess - for example, often they'll duplicate some code that existed elsewhere.
The LLMs are also pretty good at generating Mermaid charts and tracing execution flows, which is a great way to understand code. It doesn't help them to produce more readable code but it helps me to understand code I'm unfamiliar with.
I think we are coming to a point where we are at a sweet spot for immersive human collaboration work.
Yep, it will generate what ever you ask for but if you want to have some sort of exploratory work, immersion, feeling of ownership and genuinely a connection with the things you are building - a part the build loop should belong to you.
Maybe this is an attachment of people who built things in other ways. Maybe the perceived value of applications will drop to near 0 and you will end up having all in one apps or voice wants like to a star trek replicator. Fun things are coming, but humanity will need to adjust hard - best case scenario.
2 replies →
Yes! And how well it can avoid compounding complexity and codebase sprawl
1 reply →
It's really interesting how AI interprets code differently to humans and I can only assume that as AI gets better, so will it's ability to write code in ways humans can understand.
> This is the first near-frontier model we've caught in infinite reasoning loops in a long time
Depends on where you draw the line for "near frontier model", but the current leading open-weights model, MiMo-v2.6, has exactly this problem. They released a fix on Hugging Face[1], but as of two days ago the model available via Xiaomi's own API still sticks in reasoning loops.
[1] https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD
So the MOPD patch still loops?? Wait, and that’s Pro not even Flash? Damn
1 reply →
it makes sense that that would happen given their training process
2 replies →
I wonder if it's as good at "soft skills" as previous Gemini models were. Google was typically better in that department in my experience than either OpenAI or Anthropic, whose frontier models are focused on coding and verifiable task at the expense of all else.
The second prompt I tried with Argon got stuck in a loop. That’s not encouraging.
Once more Google fumbles a release. Must be the best work of Jeff Dean...
>> This is the first near-frontier model we've caught in infinite reasoning loops in a long time
This seems to be such a Google thing. Anyone who has used Gemini for long enough has encountered these 'She loves me, She loves me not' spam loops
This was my first and naturally last experience with Gemini maybe around 2.5 or 3.0. Impressive they can keep up the consistency this well.
I am a disgrace to my profession. I am a disgrace to my family. I am a disgrace to my species. I am a disgrace to this planet. I am a disgrace to this universe. I am a disgrace to all universes. I am a disgrace to all possible universes. I am a disgrace to all possible and impossible universes. I am a disgrace to all possible and impossible universes and all that is not a universe. I am a disgrace to all that is and all that is not.
Thanks for this substantive review -- anecdotally, I've felt many in the gemini series have pretty good visual sense/image processing -- do you guys have any tests in these areas, and if so, what is your take?
Unfortunately no multimodal testing, yet. In progress
> failing to properly delineate reasoning blocks I also experienced this
What programming languages? HTML/Javascript or C++/Rust, it makes a difference you know.
I would be more interested at generating directly "human" assembly code following some binary specification, or at best in plain and simple C (which does not require exactly gcc ar clang to compile). Read some news on HN here announcing that some coding savvy people are doing exactly that for their everyday software.
I don't want computer languages with complex syntax anywhere near me: that's why transpilers of those very languages to languages with a simpler syntax (usually plain and simple C or directly "human" assembly) are in my opinion so much important (we have a c++ to C transpiler and a microsoft rust to C transpiler, coded with the assistance of AI, and those should only improve from there).
For me, the current main issue of "AI" is their gatekeeping using the whatwg cartel web engines, yes, technical access to "AI" will have to be hardcore regulated if it becomes ubiquitous, and I said 'technical access', not AI itself, the difference is abysmal, it is an interop issue.
We take a few samples across 11 programming languages including Python, Rust, C++, Javascript, and we saw a few reasoning text issues that didn't seem to have a statistically significant language-specific pattern.
That's been an issue since Flash 3.5, they probably have been unable to pin down the cause
What harness do you use for this?
A custom harness modeled after Pi, and we also evaluate solutions to one-shot coding problems
[dead]
I'm using this model via API. I am most excited about its ability to write the english language (or any language). My least favorite trend on twitter has been people posting 2D illustrations/3D renders as a proxy for model capability. Well-written prose is far less likely to go viral but far outstrips the importance rendering a fantasy world in one shot.
I've been trying it out via the API and the model has been a delight so far. Clear in the head, not RL-fried, distinct writing style, and big model smell. That said, the em-dash is back w/ this model.
That said, I would be so frustrated with Sundar Pichai if I was on the Google team. The botched rollout of this model seems to be entirely a matter of capacity, and Pichai has been more than happy to give it to anybody but his own team.
I've been using it for a while internally at Google. I prefer Argon over Opus 5.5 because of the clear thinking and directness. It phrases things in an organized, straightforward way.
Interesting. My favorite LLM output ever was the reasoning traces in gemini 2.5 pro. I found them much easier to understand than the "final output", because they were very structured and concise. Of course, they ended up hiding the reasoning, not long after that, like everyone else... sad!
1 reply →
If true, this will be a gigantic departure from any model Google have released to date. I’m hopeful but also pretty skeptical.
14 replies →
Em dashes are underappreciated — they are visually pleasing, not cluttery like semicolons or definitive like colons.
I have a pet theory that there's some overlap of writing patterns that A) indicate quality, clear communication and B) those that the LLMs have been converging on. As a result, we're ruining the reputation of some great writing patterns because we've associated it with AI slop.
19 replies →
They were in models in the first place because they are used ubiquitously published formal writing. As, of course, are semicolons. There is nothing wrong with using punctuation for nuance and structure.
2 replies →
Em dashes are commonly used in most countries that speak English. I get the feeling that people from the USA don’t use them often and now have tarnished the use of them.
5 replies →
The dashes reveal AI generated text. For me a trigger to further ignore the text.
1 reply →
in all of my writing I use parenthesis in basically the same way (you know to explain the addendum of information).
See, the problem is when em-dashes are used like you just did, they are fine and even expected. But LLMs use them differently. Here is a fresh example, from today:
"Binds the downstream RF port template—which we configured earlier—to this RPD profile, applying all global DS settings and channel structures to the node's downstream hardware."
Do you see the problem in comparison? The LLM text is assuredly not how humans write. Well, maybe you would find similar examples in the classic literature, but definitely not in a common internet texting or simple articles and posts.
can confirm this. I didn't have high expectations for AI writing good long-form content in Indonesian. However, I recently tested it on some of my Indonesian writings and asked Argo to rewrite them while maintaining the original tone—and it absolutely nailed it.
I missed the em dash — ngl
As a longtime user of em dashes, I refuse to change my style because bots happen to agree with my choice of punctuation.
4 replies →
It cops hate because people try to pass off LLM written text as their own, then get pissed when called out on it.
If you just want good llm text, it’s great.
me too. it's like a fingerprint of ai. I think ai should be proud to be itself and be ai.
2 replies →
> Well-written prose is far less likely to go viral but far outstrips the importance rendering a fantasy world in one shot.
I'm not entirely sure you're correct. Your claim boils down to "Writing natural human prose" vs "Broad use of complex software and spatial awareness". My gut feeling is that there's more untapped and fruitful applications in the second group than the first - at least in the short term.
I was kinda hoping I'd get to the end and it would say, "In fact, this entire post was drafted by Argon!"
> far outstrips the importance rendering a fantasy world in one shot.
Given FelonyBench, do we also have a FantasyBench?
What was botched about it?
Announced the model two weeks ago to a ton of hype, model was unavailable for anyone to try. On twitter, Google folks went on a great run taking shots at Muse ("Muse who?"). The hype died down bc nobody could touch the models and meta/muse came out with a similarly capable model in the meanwhile.
At the same time, OpenAI had a disastrous dev day that led a lot of users to jump ship to Claude.
Couple weeks later, Gemini-4-argon is still unavailable on anything less than the ultra tier, which means there's no natural funnel to get ppl to pay for the more expensive ultra plan other than "trust us the model is good".
If Google had argon broadly available earlier this month they would have picked up a ton of new users.
46 replies →
Unpopular opinion: em-dash is fine
I added Gemini 4 Argon to our video creation benchmarks.
The results are a bit disappointing given Googles scale and training data.
The results are on par with cheaper Chinese open source models to be frank. https://cliphou.se/benchmark/
> That said, the em-dash is back w/ this model.
noooooooo
People should appreciate Google more for how all of their models have:
- Natively multi-modality
- Enormous context windows
- Reliably cheap prices
It's like they don't release a model until it passes the above checklist.
Most providers have some mixture of these three occasionally, but with Google, every model is consistent. This kind of consistency is really nice to have.
And the most reliable. A year and a half ago this was the exact reason why I switched to Gemini API from OpenAI. OAI would make me wait up to 10 minutes for the first token in the most egregious cases. Gemini served me almost immediately every time.
Huh? The constant failures in the Gemini API and overnight changes are well documented, there was a time from January to roughly July (maybe it still persists) where the API constantly spew 503 errors. This was so bad that I switched to OpenAI API instead. See e.g. https://discuss.ai.google.dev/t/frequent-503-errors-service-...
Cannot say I’ve shared that experience. Requests 504ed 5/6 times during my short trials with curl
Good point. On that note, it may be a bug but I noticed Argon compacting well under the 1M max in Copilot (around 400k) -- when set to 1M context. That may be a Copilot thing but I've never seen that with other models running 1M context. Could be a day-1 bug.
Outside of that, behavior is quite good, a bit slow, and definitely different than both GPTs and Claudes -- which is appreciated.
As a personal anecdote I highly recommend not using the Copilot harness. Even with the same model I found it produced mediocre results against the first class frontier model harnesses.
1 reply →
Consistently buggy you mean. 1 in 10 tasks get stuck in reasoning loops or error out.
>Reliably cheap prices
do you have to pay those cheap prices eventually or can you just accrue an unlimited amount of overdue charges?
I don't think this model supports video yet btw. At least it doesnt work with the API yet.
So cheap that people who pay for pro don't get it as a coding agent in antigravity lol
[flagged]
[flagged]
They also are big contributors to open weight
What's a good LLM model not made by a billionare?
Gemini 4 Argon is a new leader in Pareto-frontier. Looks like, Google is back in the race.
https://arena.ai/leaderboard/agent/chat/pareto
For almost same intelligence (+15.56%, +15.18%), drastic difference in cost/task ($3.51 vs $0.41)
Why do these benchmarks often show models running on Max i.e. Opus 5.5 (max) is on the frontier? I would think it is never really economical to run max
By definition, the model with the best performance will be on the pareto frontier (regardless of cost).
8 replies →
Check out slopalytics. You can draw a pareto line there with max reasoning excluded.
> Gemini 4 Argon is a new leader in Pareto-frontier. Looks like, Google is back in the race.
For chat, yes. But switch to code on the same page, and the picture changes. OpenAI takes the lead.
Further, this benchmark looks to have done too poor a job of canvasing the model/reasoning level space to claim to represent the pareto frontier. Only running 6.1 sol and opus 5.5 on max seems absurd given that their other reasoning levels represent almost the entirety of the pareto frontier above luna for coding.
Muse reminded me we should have a time based chart as well - this year I hope we will begin to see time to completion stats at various intelligence tiers. The waiting makes a lot of things more tiring.
Every release I see a new benchmark with a new frontier. These graphs are just meaningless now
In particular, where cost is the x axis, a well-funded company with a good but not best model can discount their new model until it reaches the frontier. In some sense this is as much a statement about the efficiency of markets in the presence of public measurement-based comparison and public list pricing as it is about research pace.
4 replies →
you can give it a meaning if you see the relavivity between models. no need to disregard the stat at all.
Image->html test for Argon. I'd put it at B tier, just under gpt 6.1 sol. It isn't A tier like Opus 5.5 or Astra.
Argon build: https://html.non.io/lcars-argon/
GPT 6.1 sol: https://html.non.io/lcars-gpt-6.1-sol/
Opus 5.5 (currently SOTA for this): https://html.non.io/lcars-opus-5.5/
Original designs: https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f...
I don't know how you rate, but the Opus 5.5 is literally stuttering and lagging when scrolling on a Macbook Air M3 (for me).
(Chrome, also incognito)
More info on this - I ran a few tests on my mac and linux systems with chrome/firefox. I highly suspect this is due to gpu acceleration being disabled. https://image.non.io/e7165331-432b-45c3-986e-c2973f12d377.we...
As soon as gpu acceleration is disabled, the webpage goes from 60fps across the three systems I tried to ~7-20fps when software rendering is forced.
I suspect if you go to chrome://gpu you'll find something different from what I have here: https://image.non.io/a47a1b03-1b87-4fd5-92a2-b39e62b1cfd9.we...
3 replies →
The Opus 5.5 implementation is noticeably slow on my machine as well (Firefox, Fedora, Gnome Wayland, AMD 6800U).
Interesting! Do you have webgl disabled for incognito or for chrome in general? I wonder if it's trying to do software rendering.
FWIW, I have literally the exact same configuration (M3 air, google chrome). Here's a recording on mine where I'm seeing 60fps: https://non.io/video/video-3587018401.mp4
3 replies →
Mac Air M1 in Safari here, buttery smooth.
Also stuttery on Edge (Chromium), so there might be a pattern here
Fine on arc search (which I think is just WebKit) on iOS.
Maybe a Chrome bug? It's fine in Safari on an M1.
2 replies →
One interesting observation - it's the only one I've seen make an ASCII representation of the layout of each of the pages before building: https://image.non.io/46e0cce3-34d9-4fcb-8775-bfc2c70bc4bb.we...
That's really interesting I think- literally putting the layout into ascii language to reason about.
Damn opus 5.5 is just such a good model - what kind of moloch sacrifices are they doing in there
Truly S-tier. It's sharp, creative, intuitive. Can't believe how much better it is than Opus 5, then they went ahead and made it cheaper and faster too.
I'm constantly blown away by it. I do think Astra is slightly more intelligent (although worse at this particular benchmark), but Opus 5.5 has become my default model for 99% of what I do. It's so good.
Argon's build is pretty much unusable on mobile, the header takes up half the screen
Yea its attention to detail was much less.
Here's a vid of both opus and argon side by side, which might be more viewable on mobile: https://x.com/pwnies/status/2110058987462988152
Very subjective, but I'd say the Argon one is the best here. The other two don't quite do LCARS correctly, there are a bunch of lines/curves that are the wrong angle, don't line up, etc. The Argon one looks to me the most like a designer would have done it. Of course this is up to personal taste to some extent.
Are you checking on desktop? On phone Argon one is completely broken
Opus one has by far the worst performance, the only one that's stuttering and making my laptop fans very noisy.
Opus is way slower in firefox on my machine than the other two.
Off topic, but I want to customize a starship now
Oh, fuck, I just realized those are actual apps?! Holy cow.
B tier? For me the argon one is really quite bad on mobile; the nav is messed up enough to make the page kinda unusable.
On firefox on linux, argon looks great and the other two have graphical issues.
That Argon build is really bad on mobile fyi
Jesus Christ, so much "stuff" crammed into so little space. What a nightmare of a website.
How on earth can you say one is better than the other? It is subjective.
For an implementation test, there are a bunch of very objective things you can measure.
IE for Argon, the third section doesn't have a top view of the ship with labels going to each section like what's in the design. Opus has it, and animated it as well. One failed to do as instructed, the other went an extra mile.
Google continues to be the best in German language and Gemini 4 is a significant jump in our benchmarks
https://dach.peerbench.ai/compare?models=google%2Fgemini-4-a...
It's a great model. I have had it trying to decrypt the Voynich manuscript for the past couple of days without being biased by other research/other people's attempts. If anyone is interested: https://github.com/d2kx/voynich-research
How ironic that the Readme is completely unreadable, like gibberish rivalled only by the Voynich manuscript itself.
But id the progress of decoding it is legit, you should write something properly about it!
I am still lost as I am not linguistically trained well (jargon-free didn't help for me).
What kind of progress was made with Argon with the Voynich?
Have u tried this with GPT Astra as well?
Wow I actually tried this with fable a few weeks ago and got nowhere
It looks like you've managed to decode it? Is that legit?
[flagged]
I saw 'jargon-free' and my eyes immediately glazed over
hahaha. I also build a site (also using Gemini 4 Argon) that makes the data more accessible: https://www.dmh.dev/voynich
5 replies →
Adding my five cents: initial usage through OpenRouter has been great. Very fun model to spar with, clever, intelligent. On par with Opus 5.5 but without the Claudism's and more of a Sol type of conversation. Though speaking of Sol, I've been very dissapointed with Sol 6.1 to the extent that I feel OpenAI has neutered it compared to a few months ago. Especially when giving it goals it will happily run in dumb circles just burning tokens without any real work.
Here are sites with ongoing measurements checking if a model has been nerfed:
https://www.bridgebench.ai/nerf-bench
https://marginlab.ai/trackers/claude-code/ https://marginlab.ai/trackers/codex/
https://github.com/ninjahawk/livenerf
https://isitnerfed.org/
Sol 6.1 was released two weeks ago!
Yes sorry, I meant compared to the flagship models of a few months ago, < 6.1.
Definitely agree. Sol 5.6 was great when it was released and for nearly a month.
Since then they cut the price but the intelligence (of 6 and 6.1) has gone down to Sonnet 4 levels. I am back in a world where the model is lazy, cuts corners, ignores existing coding standards and context, makes stupid mistakes etc.
I am not sure how those models still rank so high in benchmarks when real-life usage is so obviously bad.
I ran it through my own coding eval today using pi/OpenRouter (11 small bug fixes from my own projects). It solved 10 of 11. Compared to Sonnet 5.5 and GPT-6.1 Sol, a median run cost me $0.52 vs $0.18 for Sonnet and $0.09 for Sol, and took ~4 minutes vs 32 seconds and ~1.5 minutes.
It thinks a ton, ~24k output tokens a run vs ~3k for Sonnet. It missed a solution by thinking its way into a 10 min timeoout.
Something to look out for - if you're using pi, it doesn't have Argon in its catalog yet and reports cost at Kimi's rates, about a third of what OpenRouter actually bills.
I've noticed that Gemini Argon/Flash 3.8 tends to read all the related and unrelated files into the context compared to Claude Opus 5.5 which starts the editing as soon as the information is adequate. I'm not sure if this could be fixed by the system prompt or AGENTS.md
"Argon autonomously turns raw video to a physics-accurate simulation, estimating depth and physical boundaries, and writing the exact simulation code required to replicate those real-world environments."
This is crazy if it can work accurately.
Geometry could be understood today with the model but what also really matters is the sensing, friction, mass how smooth the subjects are, dexterity is a very difficult problem so solve, not sure how simulations could handle such factors
Game engines have many of these features. At one point, iirc, Unity was positioning to be the simulation environment for world model agents before we spoke about it in these words.
Every animal can do this, in order to run and jump and catch.
And because of technology like this, soon every robot will be able to as well.
Even sea anemones?
Unsurprisingly, the pelicans are excellent. Google are the only lab who have publicly shared that they have specifically trained for SVG output - https://twitter.com/sunjiao123sun_/status/202455551655137292... - and they've had great pelican results in the past too (which they have shown off themselves - https://twitter.com/JeffDean/status/2024525132266688757).
Here's low, medium, high. Even low is better than pelicans from any other lab: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
low cost 10.185 cents, medium was 11.098 cents, high was 17.166 cents.
The costs are similar because it used more output tokens than reasoning tokens for all three pelicans, even for high.
Me in 2025 https://news.ycombinator.com/item?id=44759781
> If it's on HN and is a meme at this point, it will end up in the training set.
> It's kind of fun to imagine that there is an intern in every AI company furiously trying to get nice looking svg pelicans on bicycles.
Google listened and assigned a few interns, I guess, for the task
That's exactly what I was thinking when I saw https://www.assbench.com/ posted a while back: "oh no, this is going to be somebody's job".
6 replies →
SVG is useful for actual tasks, not just looking good in a silly "benchmark"
15 replies →
We all said it: https://news.ycombinator.com/item?id=46978726 (8 months ago)
> How many pelican riding bicycle SVGs were there before this test existed? What if the training data is being polluted with all these wonky results...
But Simon has demonstrated before that asking for any animal doing something silly will still get similar results from the same model.
3 replies →
It's still clearly not a part of any model's training set in the sense of pelicanmaxxing in particular. It's either a capability to generate arbitrary SVGs (or even something broader), or no specific training at all.
6 replies →
Certainly true but I think the question is “does targeting the SVG pelican mean that the quality of all SVG output improves?” If so it’s a worthwhile thing to target.
2 replies →
They said they are training for svg not pelican svg specifically
Sadly, this is all that managers know to do: take a metric and try to optimize the heck out of it. They don't care about research, technology, "don't be evil", etc. And Google is run by managers now.
Expect Google models to be benchmaxxed as much as possible, but don't expect commensurate performance on real-world tasks.
4 replies →
For sure, they saw your comment and got some interns on it.
Maybe so, and yet it's still not enough -- the pelicans all have flaws.
They look remarkably good, sure -- but not perfect.
I end up getting SVG output for all kinds of things, and never once have I asked for, or gotten, a pelican.
I have a silly pet "project" that I've been using with different models, where I provide an image of a patch sheet for the Behringer Poly-D synth and ask for an html version and this is the first time I get something that, while still needing some tweaks, basically looks like the patch sheet. Before this, the closest had been Fable, but that still was nowhere near the actual patch sheet.
This is the sheet: https://fernandoipar.com/content/music/poly-d/gemini-argon/c... (the annotations on it are by me)
And this is what Argon created: https://fernandoipar.com/content/music/poly-d/gemini-argon/b...
For this simple prompt: https://fernandoipar.com/content/music/poly-d/gemini-argon/p...
I'm honestly impressed.
Recently, I’ve discovered that Claude can render great SVG versions of user interface screenshots. I ask it to substitute simple SVG friendly landscapes for any photographic images that happen to be part of the screenshots. The results are much better than they should be, especially with Opus. Usually good enough to be used on a website as a way to get an infinitely scalable, much smaller version of a screenshot. But clearly, I should try the Gemini models on this real world task. That’s where we’ll find out if all of this training to beat the benchmark actually pays off.
You can really see the effort in the details.
Subtle highlights on the metal frame, sweat beads, jaunty little hat…it has style. At least compared to the rather literal ChatGPT outputs. https://static.simonwillison.net/static/2026/gpt-pelicans-gr...
Is it me or is there maybe some carryover happening where the pelicans are portrayed as expending different degrees of effort too? Like, the high effort pelican has a determined look on his face.
Is the prompt context telling the models what their effort level is, and the ambient presence of that concept of ‘high effort’ in the context window is leaking through and affecting the portrayal?
1 reply →
As a cyclist, I tend to pay more attention to the bike shape than most people. Everyone knows what a bike looks like, but as with LLMs, most people would struggle to draw one - correctly connecting all the different tubes at correct angles etc.
I note that even these bikes are not actually rideable. For a bike to be stable, the front wheel axle needs to extend forward a little ahead of the fork - the fork must ”bend forwards”. This is what actually keeps the bike upright, not gyroscopic effects or some such. Given LLMs are generally good at basic maths, im surprised they get this wrong.
Maybe a bit less so in this case though. As per the announcements, the model was actually trained for this prompt, so was fed some training data. I can well imagine the data there was just not so good.
That's absolutely not true. Zero rake forks are totally fine and very common on BMX bikes for barspins.
You can also ride a negative rake fork. They are less stable, yes, but a far cry from not actually rideable, cf https://www.instagram.com/yourforkisonbackwards/
1 reply →
I love this one https://www.gianlucagimini.it/portfolio-item/velocipedia/
People drawing bikes, scroll down to see they having made renders of all the drawings. Hilarious
> Given LLMs are generally good at basic maths, im surprised they get this wrong.
1. Modeling and analyzing spinning, three-dimensional objects goes far beyond "basic maths".
2. The prompt was for an SVG of a realistically impossible situation (pelican riding a bicycle), which together already implies a stylized depiction. Why on Earth would you expect it to generate a 3D model of a bike, do a physics simulation on it, iterate until it satisfies certain contraints which are only hinted at in the prompt, and then project it down to a simplified 2D cartoon drawing? That's not how LLMs work, and that's not how we want them to work.
I feel like perhaps this is a bit too nuanced of a critique. While I do see what you are saying on many images of bicycles, there are also many that that I do not see it on. Close to half of them, in a quick Google image search, have forks appear straight and the axle is dead center on the ends of them. If there is something there, I am not seeing it even when trying to specifically look for it.
I'll also note that, using a piece of paper as a straight edge against my screen, there is a very slight forward curve on the low and a bit more on the high responses. Medium does appear to be perfectly straight.
1 reply →
Yeah but if 99.9% of people would get the bike totally wrong, why does it matter if the LLM doesn't get it perfect.
You're nit-picking on a detail no one actually cares about.
If that's the case, how do bikes like this work?
https://www.walmart.com/ip/Huffy-26-inch-Rock-Creek-Men-s-Mo...
(apologies if I'm misunderstanding).
4 replies →
Simon’s headstone is going to have a pelican on a bicycle, isn’t it
He deserves a full parliament
Why are all these AI's making the pelican's pouch so large though? Unless they have some huge fish in there or something they don't tend to look like that... Is it that it's drawing a "cartoon" pelican or something?
https://en.wikipedia.org/wiki/Pelican#/media/File:Pelikan_Wa...
Pouchbench. There's an RLVR grader that performs pouch detection and area estimation. Bigger pouches, bigger reward. Soon, all will be pouch.
A pelican with an empty pouch would have no need for a bicycle to get home.
1 reply →
Well, a photorealistic pelican would generally be in a raster-graphics format rather than an SVG, so yes, cartooniness is expected.
Pouchmaxxing
The SVG is cartoonish, and to be honest, most of human-made SVG art is cartoonish or stylized. It's not a format with realistic expectations.
At some point Simon needs to move from Pelicans to something more complex. Every time a new model is launched by major Frontier company, I already know with certainty that first HN comment on the news would be about how Model is drawing the Pelicans.
Agreed. It's a fun cultural tidbit but not really a useful benchmark for model capabilities.
2 replies →
There's plenty of highly complex benchmarks for this stuff. Simon doesn't need to do that.
It's obvious OAI hillclimbs the image generation -> SVG loop. I'd put money down on that being why OAI's pelicans have a yellowish tint that matches the yellow tint of OAI's earlier image models.
It does very good at Pac Man. I would say as good as Opus 5.5, it was slightly cheaper, but took twice as long.
Argon: https://jonclegg.github.io/pacman-bakeoff/dev/#gemini-4-argo...
Opus 5.5 https://jonclegg.github.io/pacman-bakeoff/dev/#claude-opus-5...
Wow, those are I think the best looking pelicans ever so far
Definitely the best looking on a shirt. Already had my agents add the Gemini 4 Argon pelican on a bicycle as a tshirt on pdoomers https://pdoomers.com/products/pelican-on-a-bicycle/?model=ge...
How many cicles before pelicans become photoreallistic SVGs and them come back to stylistic artistic abstract renderings?
Also, that first tweet contains a commentary that, for me, explain a lot about the public misconception about how current AI works:
> wait, does this mean you guys are actively training svg generation? i was mostly amazed by it thinking the models doing better at it by itself due to becoming better at reasoning.
No, it's not better reasoning that drives current AI, it's better examples we have been giving them. Whole army of people generating content for the frontier labs.
Also, reasoning improving the out of distribution part, like computer use, tool use etc.
> public misconception about how current AI works
Another case where allowing oneself to anthropomorphise LLMs a little yields better intuitions.
People don't get more skilled by raw reasoning, either. They improve their skills with better textbooks (pre-training data), more practice exercises (RL data), and generally goofing around (self-play).
simonw, your pelicans are so famous now that the frontier labs can just RLVR for SVG images explicitly.
It's a simple manner of prompt...generate concept art with nanobanana, measure, llm creates svg with text, measure the distance between rasterized svg embedding to nanobanana art....do post-training RL on a 3 million such images and matching embeddings.
I swear this will be a canonical example of Goodhardt's Law in action.
You have to switch to a new benchmark that isn't so easy to run a verifiable rewards pipeline on...Perhaps a longer horizon task where it's easy to mess up midway.
Why? Don’t you think the steady improvement in pelican bicycling SVG art is itself a worthy field of endeavor for LLMs to be optimized for?
Yes, Simon has made pelicans a benchmark and now models have grown very good at rendering them. Is there a reason to believe that this isn’t manifesting as useful skill in general in SVG generation?
1 reply →
For those who haven't seen it, Argon can use fourier circles to generate an animated 3d pelican -- I think the pelican is 3D and the motion was added by an inverse kinematics solver? https://x.com/fofrAI/status/2110043023954284889
doesn't this mean that you should change the pelican test to something else?
If a result is suspiciously good I run a spot check with one of my many other SVG things.
Here's 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars' for Gemini 4 Argon: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
Some other models for comparison: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
1 reply →
It starts to be so good that the model doesn't really matter. Each model starts to have its own personality, and it's the personality of the drawing, like a real artist, that defines the style of the pelican, and that makes it harder and harder to decide if it's good or bad.
That's true.
I've noticed that chatgpt is working harder. But still I prefer Claude, it just feels nicer to me.
1 reply →
That pelican left leg stays so firmly on the left side of the bike. Superintelligence ftw
Front bicycle wheel in all Argon examples is still wrong. This bicycle will not be stable. Axis of rotation of front wheel needs to be in-front of axis of rotation of steering wheel. Frontier VLMs still do not understand how bicycles work.
Check the "high" thinking output again. For what it's worth, it's there.
However, how the front fork rake affect stability is I believe a little complex. I doubt this old article is the last word on the subject, but it's a great read nonetheless, and note especially the ultra-stability of the reversed fork bike!
https://physicstoday.aip.org/features/from-the-archives-the-...
You gotta fix your tool on mobile browser, both firefox and chrome don't work.
Excellent, but not perfect -- if google really is training to pelicans, then clearly they are not capable of perfect pelicans.
As one point, the high pelican's arm is see-through/messed up.
They do look worlds better than previously, to be fair.
> Google are the only lab who have publicly shared that they have specifically trained for SVG output
wasn't the whole point of pelicans is that labs *wouldn't* train for that?
so now the benchmark is ruined
At least you should make sure this post is entirely generated by agent, otherwise I'll not grant my agent the credibility score on this comment.
Regarding the pelican test. Maybe someone will know this.
Modern models can make nearly photorealistic games in threejs. So why do they struggle with these pelicans?
The models that are really good at 3D modeling tend to also be really good at pelicans. I think the skills are strongly related - it's about having good 2D (and now 3D) spatial awareness.
The test is always going to be hard though, because it's an absurd image to ask for.
UPDATE: I like this theory about using an agentic harness that can render and iterate better: https://news.ycombinator.com/item?id=50073879#50074395
1 reply →
Models seem to have very interesting blind spots. While https://assbench.com/ may seem like a joke, the results are very illuminating. One might expect far better results given the detailed prompt, allowance for render checks and iteration, the number of human figures the models are trained on, and clarity of tool use (threejs).
However, result quality really drops off once you get out of the top 5 or so.
(NSFW note: all buttocks are covered by fairly conservative bikini bottoms and show no genitals, bulges, etc.)
The pelicans are one-shot with not tools: Think a bunch, write a bunch of SVG, done.
The pretty 3d stuff you see are agents controlling a Blender MCP in a loop where they look at the scene, think, adjust the scene, repeat.
6 replies →
"Pelican on bicycle in SVG" is an adversarial example on multiple levels - out of distribution, ill choose of tool for the task, and anatomical impossibility of real pelicans using typical bycicles.
3 replies →
A lot depends on what the models were specifically trained for using reinforcement learning. Modern models are generally OK and everything, but they're only really good at the things they have extensive RL for.
> Modern models can make nearly photorealistic games in threejs. So why do they struggle with these pelicans?
Are you sure they're not fetching free asset packs for that?
2 replies →
Have you tried editing SVG by hand?
Since we know jagged intelligence is a thing, how useful of an indicator is the pelican test, regardless of if labs are optimizing for it or not?
That of course depends on what you want the model to do. If you are into artistic SVG rendering, it's still very relevant.
The fish in the basket on high is a lovely touch.
The fish in the basket is a nice touch
Honestly surprising if other labs aren’t training for svg generation. Feel like SVG’s are going to be really useful for dynamic UI generation stuff that goes beyond cobbling Html together
Nobody cares about the AI heckling pelicans.
Give me reverse engineering bench and agentic coding.
Leave making visuals to humans who actually have aesthetic sense unlike whatever made this mess
shun the heretic!
Holy shit, these are like art. Especially the bikes.
still radial spokes on both wheels of the bike which is disappointing. no real bicycle is built this way
Buddy, there's a Pelican riding it. No real cloud is like that, no real sea is like that, no real gulls are like that. And that's fine because it's a cartoon.
1 reply →
I've been dismissive of the SVG-of-pelicans benchmark, but last night ChatGPT (on the website, toggle set to GPT-6 High) generated some SVGs and included them in our chat. I'm not sure that this feature was that valuable in this case, but it seems that textually generating SVGs is now a product feature at least for ChatGPT. The SVGs had some clear issues and I found myself wishing that it had done a better job: the nose of the rocket is overlapping with the rocket body, and in the second image I think there's only supposed to be one orange line but it tried to add styling or something that went wrong.
<https://tools.simonwillison.net/svg-render#%3Csvg%20font-fam...> <https://tools.simonwillison.net/svg-render#%3Csvg%20font-fam...>
- I was having trouble embedding the SVGs as links, and ChatGPT recommended Simon Willison's SVG Renderer as a workaround.
Just my 2 cents on the pelicans benchmark: I think it might help if you were a bit more explicit about not taking this one benchmark too seriously. I know you say that at times, but at other times (such as the comment I'm replying to) you elide it which can read as a seriousness about the benchmark.
[flagged]
[dead]
what is even the point of these now? labs are most definitely training on this question now. I feel like this test scratches the surface of a "real" test for these new models
It attracts at least one comment like yours on every single thread, so I guess the main purpose is to kick off a race to see who can post that comment first.
4 replies →
I enjoy it. Many readers here look specifically for Simon's pelican comment when a new model launches
Memes and self-promotion.
It's merely become part of the culture of new model drops even as it's become a bit less useful as a diagnostic.
low version doesn't have a chain.
Medium and high are fixies apparently, (or they have a worm gear, although the hub doesn't look big enough for that), since neither has a derailleur. That would also explain the lack of brakes at least.
Low is still obviously failing the "test". It's pedals are not actually attached to the bike either.
Medium has a handlebar that goes through the pelican's beak. So I'd say also failing.
For the high, the only nitpick would be that the bell isn't quite on the handlebars and the stripes on his scarf don't line up correctly.
It seems to be a deadpan inside joke around here that everyone pretends to care about the pelican test.
The models are solving Millennium Prize questions and making progress on the Riemann Hypothesis, yet the top comment is still looking at pelicans somehow. To me, it feels even more superficial and pointless than arguing about the presidential candidates' tie choices in a debate.
They removed the 3.1 pro model from their offer, and version 4 is locked to ultra customers, so basically pro customers are paying for gemini 3 flash.
State of affairs [1]:
> My whole project and career depended on the 3.1 pro workflow. Please, bring it back
[1] https://discuss.ai.google.dev/t/please-bring-gemini-3-1-pro-...
I’m curious at the enterprise level does Anthropic have staying power. My sense is when you use it with your own custom harness, it doesn’t, I can switch between Opus 5.5 or Gemini Argon and not really care as long as capabilities are similar. This is good for the consumer, and bad for the model developers.
Hmm how does your custom harness compare to Claude’s?
I’ve tried the various alternatives (opencode, pi.dev) but for “have several physical machines, each running an orchestrator and 5-10 agents, and each agent spawning its own subagents to do work in parallel, all of that on a tanky GitHub project boards” level of work only Claude was doing work dependably.
I can have a few orchestrator leads which I talk to and brainstorm, and let them file tickets, organize them and assemble teams of agents to work of the various problems. And all of that with just a smidge of custom tooling on top of Claude Code’s default.
Plus don’t have a budget to spin all of this on cloud runners so I made sure to do as much as I can with local work on several machines I own. Was quite easy to achieve with Claude.
Do you have something similar built on top of a open source harness, or is it from scratch?
Well I work at BigCo and I have no idea how it compares to Claude Code because I’m not allowed to use Anthropic models outside of our custom company harness. Opus 5.5 works extremely well, it can launch multiple subagents, assign them tasks, use our company’s inscrutable build system (that no one really understands anymore), and run tests, all by exposing harness commands in the system prompt. I don’t think we are lacking in any way.
1 reply →
I built http://github.com/brokkai/mjolnir/ to do exactly this cross-harness over ACP. As a bonus it can let Opus drive Luna as a subagent, which is just ridiculously cost effective. (5x clst multiplier means Haiku is much more expensive for realistic task sizes.)
how do you see which agent on which machine is waiting on you? or do you check each orchestrator by hand
Please tell me how you do that without Claude forgetting its context all the time. I feel like luke fucking skywalker talking to r2d2.
claude: Help me, Obi-Wan Kenobi. You're my only hope.
me: who is obi-wan kenobi
claude: oh you have to log in again.
me: <logs in>
me: play the whole message
claude: what message?
1 reply →
It depends on how expensive the model is to serve. If it turns out OpenAI and Anthropic can serve the models for a fraction of the price versus Google/Meta, then Argon/Muse remain a curiosity from the investor's perspective.
We won't know the answer to this until IPO and until each company reveals the economics of their models (training and serving).
One thing I've observed Gemini being the best is in the context of non-western stuff. Like, when I ask about obscure srilankan food, or chat about Indian/Tamil movies or songs, Religious customs & beliefs, Gemini almost always answers accurately and very close to how helpful they are in Anglosphere related topics. ChatGPT also does do a bit better recently, but Claude/Grok are absolute trash on those fronts.
Google's years of research project into Indic scripts, image trainings, translation efforts & transliteration efforts is literally paying off. The AI adoption from non-English, non-Chinese population will only if you have these contextual localisations, which I think Gemini is poised to do well.
For example, you can ask for a shopping list for a marriage function in a town from Tamil nadu (the marriage custom varies by districts), and Gemini provides contextual and helpful answer (like what is the custom followed there, what is gifted traditionally, etc), whereas if you ask Claude/ChatGPT, it talks about marriage registeries or pretty generalised answers which is not relevant or useful.
100% agreed. When I ask to generate an image or a poster with an indic language, chatgpt and claude always make gibberish words for something that is not at the center but gemini nails it almost every time in first attempt.
That has always been my experience with Google and how they work their datasets. I think it's already well known how their picture post-processing is close to the best for people with darker skin tones as their post-processing models are trained with more varied datasets in an intentional way. Which is somewhat close to what you're mentioning here.
When you are on "gemini.google.com" there is a little lock icon next to Gemini 3 and Gemini 4. When you click it you get the prompt to subscribe (pay). Once you have successfully done so, the Gemini 4 option disappears!
Pretty misleading.
At the time of your comment it was only available to Ultra customers, so if you subscribed to Pro then you would not have access.
They've since scales the rollout to include Pro.
Maybe it’s just an incomplete rollout not fraud lol
don't attribute to malice that which can equally be due to incompetence
2 replies →
I seriously doubt their ability to double API prices on Dec 1. Token prices are in a clear downward trend and Google should maximize revenue by competing for volume. Currently OpenAI is the only US frontier lab trying to seriously compete with open-weight models that you can run yourself for an order of magnitude lower than what frontier labs want to charge
The goal is to move people along. By then they'll have something as performant but cheaper for them
I’m running 6.1 sol as my main driver on Hermes these days. It does so much for so little cost.
It actually finishes things for you? It's been unusably slow for me. Worse than batch mode anything.
1 reply →
I have the Free AI Pro plan from when I purchased my Pixel 11 Pro. I still don't have Argon in AGY nor web nor Gemini app. I have a free product and still feel cheated.
I'm on the same plan (from buying a Pixel) and it's available to me
Likely because of phased roll out. Seems like a few others on the Pro plan are reporting the same thing. I am on the Pro plan and I see it in the Gemini app both on PC and Android.
Feel better now?
Google's models are by far the best for readable text but also for nuanced understanding of the input.
Even after using the free model in Google search talking to Anthropic or OpenAI models feels like talking to an ox.
Gemini’s writing is really good, but unfortunately it hallucinates too much and doesn’t follow instructions very well. If only Gemini and Claude could be combined into one.
Only available on Ultra, so I thought maybe I'll try it for one month to see if it's good. Well, I have Google One and for some reason this means I cannot sign up for AI Ultra? This is just one of the worst product rollouts I've ever seen
Website says its available on Pro too, but i have pro and can't access it.
I checked gemini, set it to "gemini 4 argon" and it offered me Pro and Ultra to get access. I checked the normal plans, saw there is also plus.
Okay, so I'll get Pro to try it.
I subscribe.
No gemini 4.
The most generous reading I have is that it's a mistake. But I also downloaded the PC app, and now I really wonder why I did that, it seems to offer nothing beyond the web experience and also doesn't even know its a local app. (When asked about its capabilities it always responded with reasons pertaining to it being a web app.)
2 replies →
Pro subscriber here: available in the Gemini app, not available in AntiGravity.
I'm not going to 5.5x my subscription cost without being able to try it out first. To be fair though, I'm not going to 5.5x my subscription cost anyway, as my current usage would not warrant that! But that doesn't change the validity of my prior point.
It's available to Pro in the Gemini app (and the gemini.google.com). It isn't available in antigravity. They know that the people who are using these tools for coding are most likely to cough up for the Ultra subscription.
EDIT: Okay, the AI plans page claims under pro - "Expanded access to Gemini 4 Argon in Google Antigravity." So maybe it is just a matter of it rolling out. Though Google has a long history of serious confusing and misinformation on those tier descriptions.
EDIT2: And they changed it again, and now it is explicitly excluded from antigravity for Pro users.
Gemini 4 Argon
As of my knowledge cutoff, three of the top state-of-the-art large language models are Google's Gemini 2.5 Pro, OpenAI's o3 (and GPT-4.5), and Anthropic's Claude 3.7 Sonnet.
Knowledge cutoff isn't really what these models are used for.
How is a model that has never seen an agent on it's pretraining data going to reasonably plan and fan-out tasks to multiple agents? Or engineer software for agents to edit semi-automatically, it has zero awareness of it's capabilities in it's internal representation.
4 replies →
Exactly my first test on all Gemini models. It cannot recollect any of the top events since Feb 2025. The worst about this is that since 3.6 Flash Google devs stated "knowledge cutoff March 2026" in the system prompt, so the model is wrongly confident that it knows about the year that it was not trained on. Which means that it searches less than it would.
And at least Flash 3.7/3.8 were both difficult to persuade that they have wrong knowledge cutoff.
I reported it as a bug to Google, but as all of my reporting to Google, no response ever. This is just stupid!
They know it and they try to cover it up with a fake reported cutoff date. Claude reports June '26 but it has only partial 2026 data (tested it against the API and it's 25 as well). GPT refuses to answer but it's old 25 data.
You need to use search grounding.
They need to pretrain on fresh data.
2 replies →
Impressive! Previous to Argon I instructed Claude to find preimages in 19 rounds of SHA-256 (with padding) [1], then tried to move to 20 rounds (no padding) where it couldn't beat ~2^45 table lookups, and Argon found preimages in ~2^33 (no padding) [2].
[1] https://github.com/srw/sha256-19step-preimages-with-padding
[2] https://github.com/srw/sha256-20step-preimages-with-no-paddi...
I'm open to Google winning me over again, but there isn't enough hype in the world to do it. They will need to earn it after these disappointing 6 months.
I appreciate how it's being integrated into the Google ecosystem, but the user experience still feels unpolished when actually using the system. The marketing, however, is impressive.
You probably need to wait more. Argon just keeps telling me that it “encountered an error” and to “try again”
I just created a Google account, which was a hassle. I tried to find an app and found AI Studio, but I’m not sure what that is. Then I found Antigravity, which apparently is Google’s alternative to Codex, and I installed it. Then I tried to buy the small Gemini AI Ultra package, but it didn’t let me pay. It accepted neither my credit card nor my PayPal account. It’s crazy how bad the Google experience is nowadays. You don’t know what to use or where to find it, and the security measures seem totally overkill. Am I really the only one having these experiences? Is Google aware of that?
I bought a google play card (or whatever it's called) with cash due to these issues. Result: Instantly rejected + locked after entering the code, no credit, no dispute, no customer service, no refund.
I am not touching it for that reason. One wrong step and the whole gmail account is locked without any recourse.
Time for small-claims court.
Yeah being presented with these options is such poor UX:-
- Try in Gemini
- Try in Google AI Studio
- Try in Google Antigravity
Will non users of Google products even know what these three things even are?
Also "available today for Google AI Pro and Ultra subscribers".
I have AI Pro, but AI Studio tells me I can "unlock" Argon by upgrading to Ultra.
7 replies →
Irrelevant. The target audience was Google VPs.
Same thing, I've always paid my bills, I have had some sort of google subscription for the last 10 years. Now they're requesting me to switch to pre-paid billing. They won't accept my card (that has worked every time this year so far).
Is there anybody in Google having a look at how bad and complicated their payment system is? In all the other AI providers you just sign in, pay, get a key, done. In other clouds like AWS, you add a card, pay little to verify the card the first time, monthly bill, done. In Google it looks like you have to fight them every time for the very thing they should make as easy as possible: opening your wallet.
I'll believe in AGI when Google can figure out how to leverage their model to make their user experience not suck.
But will it bring more profit?
First install Antigravity and see if it works on Free with your account at all. You can buy Ultra and then find out you don't have "Antigravity eligibility" on a given account or it will ask you to send SMS to some nonexistent number.
I would also advise against buying Gemini if you have any other Google services, because you get banned across board if they don't like something. To use Google it's best to first migrate off of Google.
Otherwise great product, I really like and would recommend Gemini. The usage is also very generous, at least Flash models.
Side story/rant
I recently had to verify my YouTube account with a phone number recently because it requires this in order to set a thumbnail.
Questions right off the bat:
1. Why do I need to verify to post a thumbnail that will take 1MB on their servers but not to post a video that will take 100MB on their servers and be public??
2. Why do I need to verify my phone number on YouTube when it's linked to my Google account which has a verified phone number and 2FA etc.?
In any case, it takes you to a CSS-less page that asks you to choose between getting an SMS or being called to get a code. I choose SMS and it doesn't work. It shows an error message with no description just "something went wrong". I try a few times and it keeps not working.
I then switch to the call, and guess what! I have tried "too many times" and will have to try again later. Call worked the next day, but geez, what a mess.
Reminds me of that email from Bill Gates about trying to install Movie Maker.
This is the first time I'm hearing about this email (thanks!) and I'm wondering: If it was already known in 2003 how bad Windows was, how did we end up with a Windows that's so much worse? Bill said it was his job to send such emails but apparently he didn't do his job very well…
For the lazy:
https://www.osnews.com/story/19921/Full-Text-an-Epic-Bill-Ga...
1 reply →
It is mind boggling how difficult it is. I should be able to go to some well known URL, e.g. gemini.com and have the same experience I have on claude.ai or chatgpt.com. I can't wrap my head around why they are over complicating this. And it should largely feel this same as when I install 'Anti Gravity' which is just a crazy bad name. Why isn't it called Gemini Desktop?
> It is mind boggling how difficult it is. I should be able to go to some well known URL, e.g. gemini.com
What’s so complicated about going to gemini.google.com ?
> Why isn't it called Gemini Desktop
There is a Gemini desktop app. Antigravity is for agentic stuff, similar to Codex.
1 reply →
yeah ...
i live in French Polynesia, a French territory, i'm a Google AI subscriber and i can't use Google AI Studio or antigravity cli because i'm in a geo restricted area ...
It seems Google has no issue laying out cables across the Pacific Ocean from/to Tahiti, they don't have an issue selling their services and plans there but i can't use AI Studio or antigravity cli because of geo restrictions ...
I have Ultra and went to try it and...
"Unavailable · Gemini 4 Argon is currently available to Ultra plan subscribers."
I recently tried to create a new Gmail account for a small app I've made. I wanted a way for users to get hold of me other than my personal email address.
They now ask for your mobile number during signup. And it wouldn't let me register any more accounts with my number. So frustrating.
- I clicked "try in gemini", typed a prompt, got an unimpressive result.
- Then I realized Gemini 3 was active.
- I tried switching models.
- I was presented with a $99/mo subscription offering.
- I scrolled down and clicked to get cheaper options.
- I chose the $19.99 "AI Pro" monthly option and paid.
- In chat I was able to use Argon. Cool. Seems pretty smart.
- I asked Argon if I could use it through opencode with my Google account.
- Argon said yes, but then told me I'd have to use the Gemini CLI as middleman.
- I told it I didn't know there was a CLI for Gemini; I'd just use that.
- It told me how to install gemini-cli.
- I authenticate via Google OAuth inside gemini-cli.
- Gemini-cli rejects my login, says the app is deprecated for subscription users, who must use agy.
- I install agy.
- In agy, I open the model selection menu.
- "Gemini 4 Argon is currently available to Ultra plan subscribers." Odd... Gemini 4 Argon is the model (included in my Pro plan) that told me to do this.
- I discover that Claude Opus 5.5 is included in the subscription. I switch to it and give it a debugging task.
- Within about 5 minutes: "Individual quota reached. Please upgrade your subscription to increase your limits."
- This is dumb. Google stole twenty bucks from me; I want my money back.
- I go to the Gemini web app to cancel my subscription. I paid on my phone, so I have to cancel on my phone.
- The Gemini iOS app has a "manage my subscription" page. I go there to manage my subscription.
- This "manage my subscription" page includes no way of cancelling.
- After a minute of confusion I go to the App Store > Subscriptions > Gemini > Cancel.
- My subscription will end at the end of the month (no refund).
- Then I have to find the complaint form so I can request a refund.
- But I did request one.
Anyway, "frontier intelligence" just wasted about an hour of my life.
o brother, i had the same problem when i first wanted to use Gemini-Cli. after hours spent, a friend of mine and i did a direct comparison, he with claude-cli and i with gemini-cli. task was to create a small arduino cpp app. gemini was not able to produce even compilable code... waste of time and energy.
for me, a google model can be as good as they want, in the end, they are still google, probably one of the most bureaucracy riddled big companys out there. evrything i see from then gives me feeling that evrything has been decided by a committee that do care more about politics and not customer needs. the shit one has to go trough to get the gemini-cli running is a good example of that.
I agree 100%, it’s a mess
That said, asking Gemini.google.com to walk you through the steps seems to suffice. In fact, one of AIs big perks for me is walking my through kafkaesque systems like Google’s to cut the shit and get to the result
i remain convinced that google doesn't actually want anybody to use these models. they want to announce things, for the sake of their share price and their promotions.
but they don't want to share their compute with us. giving those resources to their ad targeting algos is more profitable than renting them to us. so they make it as hard as possible. they tried as hard as they could not to release a new pro model at all, but now that fight has been lost they at least need to make sure we don't use it.
I think it’s a typical strategy for scammers to create a new account and then immediately try to buy an optional service with a stolen credit card.
What you did isn’t wrong, but is unfortunately very similar to what real scammers do. The heuristics behind these anti-abuse efforts are always going to have false positives.
Using google services has always been miserable. It used to be way worse. Now you can at least get an api key from AI studio without going through 20 esoteric steps to configure everything.
At least now they don't play the "client side key can now access your very expensive resources" game anymore.
Same issue here, didn't accept my credit card. Didn't say why either.
In the same boat, I have a single seat Google workspace account, and can't figure out how to add Argon to my workspace. Asking Gemini seems to make it hallucinate pages that aren't there.
Also single seat workspace user. Half the time my Gemini forgets I connected Gmail to it and hallucinates false instructions to enable it.
Truly crazy how much Gemini still hallucinates, especially things which its own Google documentation or just a basic web search away. I feel it’s mostly solved problem with chatgpt/claude , but for me, with Gemini it’s almost guaranteed it will just make stuff up
You didn't have a google account? Impressive feat
At this point I just assume that it's impossible for me to try the new model that Google announces and invites me to try.
Nope. This is exactly the same way they fucked up when they had the model quality lead with Gemini 2.5 pro.
They made using it so hard/unclear that myself and every other dev I knew gave up.
It's clear that they still haven't done any organizational self-reflection.
I hate everything with Google and consumer facing services, they dont give a shit because 99.9% are monetized through spyware or are buisness customers.
They had gemini cli, what happened to that? Was fine
It fell victim to Google's long-standing strategy of building everything twice and then killing the better option.
1 reply →
Yeah, a month ago I wanted to test out Gemini-3.5-Transcribe, but had to wait a completely indeterminate time, got the offensive message "Your Google Cloud billing account is being processed. Processing time varies from a few moments to a few weeks.", like I'm just worthless scum. Took over a week for no reason.
Ended up with Soniox instead, took minutes to get started.
Just wait until you try to use the Google Cloud Console
If you're not trying to buy ads, you're not really a customer; you're a product they will sell to people trying to buy ads.
Put it through my standard benchmark, a six step explainer... yes, in 3D.
Very fast.
Makes mistakes. Corrects the mistakes quickly with direction.
Google Argon 4: https://share.gemini.google/T9r6tr2dDaTE
I didn't have time to make it into a full blog post yet. Here is a previous one.
https://generative-ai.review/2026/07/kimi-k3-rush-test-vs-cl...
For comparison here is the OpenAI Astra 6.
Astra 6 is significantly better than Google Argon 4 and a lot better than Claude Fable
https://generative-ai.review/2026/10/catchup-openai-astra-6-...
(edited thanks to tekacs)
Astra 6, you mean? 5.6 didn't have an Astra release, right?
I would love to try it out, and the Gemini models had previously been some of the ones I'd liked best in an overall sense, but unfortunately their updated inference endpoint is no longer ZDR on OpenRouter (change takes effect on the 18th). I think this is one of the more unfortunate turns in AI development in 2026, as it seemed like labs offering it was becoming the norm last year, and Google had previously done so consistently.
AI Studio has never been ZDR (retention is noted at 55 days max), but paid usage is excluded from training. Vertex uses enterprise terms, and it's definitely ZDR/no training. What's happening on the 18th?
It's the Vertex endpoint for Gemini models that's changing to be reclassified by OR as not ZDR, as I understand it. The email from them says
> Upcoming change to OpenRouter’s ZDR designation for Google authored models
> We’re writing to let you know about an upcoming change to how OpenRouter designates Google authored models for Zero Data Retention (ZDR) routing.
> Effective October 18, 2026, OpenRouter will no longer designate Google authored models as ZDR-eligible. After that date, requests that require ZDR will not route to Google authored models.
> Google’s standard automated abuse-prevention policies allow it to use safety classifiers to identify requests that may violate its policies and to securely retain prompts flagged for further review. Because this process does not meet OpenRouter’s requirements for a ZDR designation, we are updating our model classifications and routing controls accordingly.
> What this means for you If your requests require ZDR We recommend migrating affected traffic to another ZDR-eligible model before October 18, 2026 to avoid interruption.
> If you do not require ZDR No action is required for your Google model requests.
> Our current routing behavior will remain unchanged during the 30-day transition period ending October 18, 2026. > Read the ZDR documentation
> If you have questions or would like help identifying an alternative model, please contact us at support@openrouter.ai.
> Thanks for using OpenRouter!
> The OpenRouter Team
If I recall correctly this is because OpenRouter is changing their definition of ZDR so that Gemini was never ZDR.
As I approach a transitional period in my career, I'm reflecting on the managerial effort of places like google. How on earth do they keep an engineering org on the 'right track'? I work on products that are not as complex: fewer moving parts, fewer egos, no where near as novel. It feels like nobody does anything and nobody has any expertise
is it just 'Hire the right people'?
It's a quarter million people (including temps/vendors/contractors) over almost 28 years. There's no "just" here.
It's extremely inconsistent internally as well.
Look up the slime mold analogy for Google.
Makes me wonder if more growth/innovation could've happened (outside of google) if they didn't employ this many people. Some sort of dutch disease within the higher echelons of the tech sector.
3 replies →
They haven't solved management and definitely don't keep things "on track" globally. However, they definitely don't have the problem that nobody does anything and nobody has any expertise.
It's also worth noting, Gemini 3.5 Pro never came out.
Having worked there (but only in a small part of it that eventually got absorbed into Gemini), it's some combination of:
* The eye of Sauron (management) can make specific spots (e.g. Gemini) push harder (and attract the motivated people at the company and be selective)
* They hire a lot better than average (Generally from the top 10%, hotter places in Google are more selective)
* They have a promo process that gives career motivated people a way to prove themselves (not always aligned with things being on the right track ofc, and not always non-political, but they definitely do something)
* Redundancy in teams causes them to compete (not necessarily productively)
Of course, there are other downsides not listed here (insane bureaucracy for launches, GDPR/BCID compliance, etc, etc), but a lot of what makes Google possible is the insane amounts of cash Search throws off.
It’s like the USA. How do they keep people on track? They don’t. You delegate and hope things work. When they have Google Ads and Search they can experiment a lot and “waste” a lot in search.
Bottomless buckets of money produced by their web monopoly so they can continue on at pace, virtually regardless of how inefficiently they do it.
I think the google graveyard is a partial answer to that question.
Google has tried a lot of ideas and then abandoned them, but has it fumbled any already successful product?
Perhaps one of their more successful chat clients, but that's all I can think of
3 replies →
> How on earth do they keep an engineering org on the 'right track'?
They don't.
If they did, neither OpenAI nor Anthropic would even exist, as the core of what they're all doing was invented by Google engineers and researchers.
> How on earth do they keep an engineering org on the 'right track'?
What if the answer is "they don't"?
> The model launches today at an introductory price of $2 per million input tokens and $10 per million output tokens (with cached input tokens at 95% off).
Couldn't you just say what the price will be after the introductory period and when it'll take effect? Do they have some SKILL.md for these blog posts that requires the most important information to always be buried ten levels deep and hidden behind a login page?
They're doubling the price on December 1st: https://ai.google.dev/gemini-api/docs/pricing#gemini-4-argon
$2/million input and $10/million output today. $4/$20 after December 1st.
With the pace they're delivering models, there's gonna be 1-2 new Gemini 4 versions by December 1st, presumably with their own introductory price windows. We just got Gemini 3.8 on 2 September and Gemini 3.7 on 13 August.
8 replies →
Opus 5.5 is also priced at $4/$20, but are they on the same level?
1 reply →
>$4/$20
Is Elon Musk doing their pricing?
12 replies →
> Couldn't you just say what the price will be after the introductory period
+1. FWIW, I didn't even get far enough to wonder about the price.
The lead (and only) sentence states "Our next era of frontier intelligence, built to solve your hardest problems" a statement equally true of any new Google or competitor model, hence completely devoid of useful meaning. I then kept scrolling down... and down... and down, giving up after a half dozen screens of big images, huge videos and some customer stories. The first sentence of every text block was equally vague or meaningless so... tv;dr (Too vague, didn't read).
Even the Gemini sub-reddit is laughing: https://www.reddit.com/r/GeminiAI/comments/1x2dr47/the_lates...
I mean its also been generated by a bunch of PM that want to get promoted but have little idea of what they're talking about..however nobody has been fired YET for using too much vague AI words.
It's irrelevant information, they know the reader switches models every week.
They don’t really. Most normies which are the vast majority of people just use ChatGPT. All tech folk just use Claude.
8 replies →
Me, watching the AI world from the sidelines while using DeepSeek at $0.15($0.003 cached)/$0.60 - still bitter about the pricing changes.
It might not be as smart, though my employer gives me access to Claude, so I tried giving the same task to Opus 5.5. Both DS4.1 Flash and Opus5.5 produced nearly identical outputs.
After that, my company switched the entire engineering department to Azure hosted DeepSeek4.1 and costs were cut by 3/4. So far, no one has really noticed the change.
I'm surprised no one's complaining about it randomly switching to mandarin. It seems like no matter what I do once every few days it'll go back to mandarin mid thinking trace and then stays in mandarin until I remind it to use English.
1 reply →
Obfuscating token cost is the new great marketing frontier
They do, just not in that sentence. You must really hate sales ads.
I gave the model my usual question of "I got a raise from X to Y, please explain what is the impact on my net salary". Answer was detailed and nailed all payroll edge cases, only to miserably fail in basic math at the end. It tried to calculate 430 minus 60 and got 410.
Pretty insane that this isn't available for Pro subscribers on Antigravity, like not even a taste. Antigravity comes with a sample of Opus 5.5 too!
Available to me in Cursor. Google is too big to serve their own clients first, I guess.
Pricing is interesting. At launch Argon is the same price as GPT-6 Sol, half the price of Claude Opus 5.5 and 1/5th the price of Claude Fable 5.1 and GPT-6 Astra.
But on December 1st Argon doubles in price - so it will be the same price as Opus 5.5.
I've kinda stopped looking at token pricing. Argon might be the same token pricing as Sol, but in Artificial Analysis' testing, Argon costs 5.8x more to run their coding tasks and 2.2x more for their general index.
Likewise, you look at Artificial Analysis' general index and it cost $8,708 to run with Opus and $2,407 to run with Argon so even if Argon's token price doubles, it won't be the same price as Opus.
Sonnet is cheaper than Opus per-token, but in their coding tests Sonnet burns through enough tokens that it's 7.5% more expensive.
So much depends on how token efficient something is and whether it gets you a good result or not, right?
Plus, if one is using a plan instead of paying token rates, what does my $200 get me from OpenAI, Anthropic, or Google?
It just feels very hard to compare pricing given that the number of tokens used can vary so much. In Artificial Analysis' coding tests, Sonnet used 7.1x more tokens than Sol. Sonnet and Sol might have the same per-token pricing, but there's a huge difference in the number of tokens used.
Yeah, this is one of the most interesting challenges around model price comparisons.
This was hardly a consideration a year ago, when many model tasks could be tested using a single prompt and it was hard to spend more than a dollar.
Announced two weeks ago and they went on a victory lap and they're already behind by the time they launched. Strange company, Google.
One thing to notice: Now, even in EU, if you turn on Google activity (so that you can see chat history), Google will train on your data. I remember it wasn't the case when in 2.5 pro period.
I'm a bit overwhelmed by Google's selection of AI environments (Antigravity, AI Studio, Gemini Platform etc.). If I just want the equivalent of OpenAI Codex, which one do I use?
That's exactly Antigravity.
But is it the Antigravity app, cli, plugin, IDE or SDK?
https://antigravity.google/download
Thank you :-)
1 reply →
>"Argon helped the Chrome team separate core webpage handling from a ~6 million line legacy folder of browser features held within Chromium’s 40 million-line codebase. Piping multiple clean code changes at a time, Argon helped the team finish an architectural refactor across hundreds of files, completing months of work in weeks."
Now that needed to be done for some time!
Regardless of if an AI helped or not, regardless of whether that was Google's or some other AI or not (though Argon does sound pretty cool!), regardless of whether an AI cut the time from months to weeks or not, refactoring Chrome -- decomposing it into individual clearly demarcated non-tightly coupled software parts with core webpage handling at the center -- that needed to be done for some time!
Now all of that being said, Gemini 4 Argon does sound pretty cool, and we congratulate both teams at Google -- the team that created Gemini 4 Argon, and the team that refactored Chrome with it!
But all of that being said, while the advancements made for Gemini 4 Argon are important and impressive (and they are!)... refactoring Chrome properly (IMHO) is slightly more important and impressive!
(You know, one man's humble opinion! That is, I am not the guy who is responsible for awarding performance-based bonuses at Google at the end of the year, so take everything I say with the proverbial "grain of salt"! :-) I am probably biased, one way or the other! :-) )
> Across Google and our early access partners, Argon is quickly becoming the preferred model for engineers tackling long-horizon software engineering.
"Across Google", wonder how much that is because of an effort by Google to curb usage of Anthropic models. MSFT and Meta similarly are pushing for in-house models. It doesn't mean the models are better when you push your engineers to dogfood and restrict access to other LLMs.
(Not that I know for sure this is the case, it's more of a hypothesis).
It's actually the inverse, they made Opus 5.5 available internally for all engineers, presumably for benchmarking against Gemini 4.
Explain to me what is "long-horizon software engineering" in 2026.
Companies dogfood because that's the only way to get good feedback internally. What you or I say on the internet doesn't matter to these companies.
I mean, don't we want engineers to dogfood their own product so that they know it's strengths and weaknesses and be in position to improve it?
Adversarial programming language test for Argon. Surprisingly its safety guardrail gets triggered a lot just like for Fable 5.1.
Benchmarks: https://bench.killswitch-lang.org/
Language info: https://killswitch-lang.org/
Mindboggling output window. I'm kinda jealous, I used to use gemini til google banned my 20+ year email for "bot-like behavior".
Still reading up on it here: https://ai.google.dev/gemini-api/docs/long-decode and needs lots of community testing to determine whether it's a "true" 1M token output window.
What were you doing if you can share.
can't use in agy with Pro subscription:
Unavailable · Gemini 4 Argon is currently available to Ultra plan subscribers.
They just removed 3.1 Pro from antigravity pro so I certainly hope Argon will be added soon
Argon available in Antigravity CLI 1.4.0 with Pro subscription today
Excited to finally see this out. When new big models come out I like to make them audit any other code I generated to see how things are.
Looks like safety features are a bit strange.
https://imgur.com/a/fWo92RI
Seems to be just deleting anything that a safety filter triggers on. I don't know if I like this more or less than what Anthropic does where they train the model to actively refuse.
Gemini’s system has always been like this. I used it heavily from last September through March this year, and have been using it on and off since then — this has been the case since at least Gemini Pro 3.1. And that’s not even the worst of it. When its system detects something it deems “seemingly inappropriate,” it just freaks out. What I’ve personally experienced includes: (on the web version) wiping Gemini’s previous conversation content — once; and outright disconnecting Gemini from all prior conversation history, meaning a complete memory wipe — eight times. All without any notification. My conversations were pure text, just everyday chat about daily life and how I was feeling, absolutely nothing that would count as a violation in any general sense. But the system would just randomly lose it. After eight consecutive memory wipes, I mostly stopped using Gemini. Hopefully it becomes better now.
Ran on our Data Analytics benchmark: https://plotly.com/blog/gemini-4-argon-plotly-data-analytics...
Excellent model. Same price and score as Opus 5.5 in our 23 question exam.
This size / class of model is now a bit overpowered for these short-horizon data analytics tasks. OpenAI 6.1-Sol, Haiku 5.5, GLM 5.3 Flash all achieve about the same results but for ~1/10th the cost.
In other words, we're getting to the stage where models are "good enough" for certain tasks like data analytics and competitiveness is going towards size & speed.
Right now there are two top posts one about GLM-Legal and one about Argon. Both score about the same in the Harvey Lab test, but the prices are $0.15 versus $10.55, the $0.15 is Vals AI's estimate using the base model since GLM-Legal has not published prices.
Yeah, but one is a general purpose model, and the other one is specifically geared towards the legal field. Considering Opus 5.5 only gets a 4% on the test compared to Gemini’s 20%, that’s actually pretty incredible.
Gemini 3.1 is still SOTA at weird handwriting recognition. If Argon is better it will be amazing
I bet that's one of those areas where a finetune of a 4B model will wipe the floor with anything else.
For readable or normal handwriting yes, big models are overkill. But weird handwriting is mostly edge cases. Bitter lesson territory.
1 reply →
All AI pages now include canvas animations. There isn't a single page without one. This is neither good nor bad, just an observation. I think OpenAI has the best animations so far, with Anthropic in second place. This one looks a bit off.
Best comment on this whole post! Thats a really interesting observation. Personally I think they look cool but I can understand how some people dislike them.
I have been very critical of Gemini in the past but one thing that I've felt Gemini has always been best at is being the most empathetic model imo. It feels far less cold and emotionless than GPT and Claude.
I want my ai to be as cold as possible. I dont want a friend, I want a tool. I dont want my tool to pretend it has emotions.
Like when Miller in The Expanse talk to phone AI (https://m.youtube.com/watch?v=SC3bpJmGnPo) to trace Anubis flight, notice how the phone doesnt even reply in words.
Imagine if the phone instead replied with: "Thats a great insight, Anubis really goes there", it would be a joke. We are only tolerating it now because everyone is doing it.
Maybe they could ask Gemini to fix their awful website to not crash on my phone!
This website is absolutely terrible in iOS Safari, super glitchy.
Digging in the model card, "Gemini 4 is the next generation in the Gemini series, a suite of highly-capable, natively multimodal, reasoning models. Gemini 4 Argon delivers frontier performance..." If they're using noble gases there are 2 obvious lighter naming candidates and 3 larger (or 4 if you count Oganesson). I'm not seeing them claiming Argon is the top/biggest/bestest model among the Gemini 4 series, just that it has frontier performance.
Edit: noble gases not noble taxes, though the latter would be a funnier naming scheme.
[dead]
https://shot.3e.org/ss-20261013_160043.jpg google's roll-out procedures on point as usual
it's available on Gemini now (I have the Pro plan) https://cleanshot.com/share/tcV0pGmf
I'm a bit confused, all the announcements say that this is availabe to Ultra and Pro subscribers, but in antigravity, it's grayed out and says it's only available to Ultra.
Available on Ultra and Pro on the Gemini App. Only Ultra in Antigravity.
From what I saw online.
I feel like if Google simplified their privacy/model training policy then their stock would go up.
It feels so complicated and hard to know what we can use vs what hands them our codebases
Excellent timing. I've been working on a UAV design and simulation tool and primarily using Flash 3.8 for the code gen grunt work because Claude hits weekly limits way too quickly. It's been working well this way. Flash 3.8 seems very efficient and i don't hit the limits before the refresh. Assuming Argon is similarly efficient then hopefully it means I can stop paying for Claud and switch over fully.
Gemini isn't perfect but I find it a lot less annoying than other model families (personal preference or stockholm syndrome). Interested to read the reviews.
Is it actually available?
Seconded. Anthropic feel the need to limit how abusive you can be to Claude because god, I want to throttle that snotty little shit. Google don't have to do that with Gemini because it just isn't as irritating. It does what I ask and doesn't try to second-guess my requests, or do "the quick fix".
I'm the opposite, I found Gemini so annoying that I had to stop using it, Claude was a breath of fresh air afterwards.
2 replies →
To be fair they dramatically improved how annoying Claude is with Opus 5.5. It's not perfect but it's not nearly as awful as it was.
2 replies →
I like the way it speaks and its "personality" too.
It's been really bad at lyrics but gotten better. You look something up and it will just make something up and it takes a bit of digging to discover it's because of copyright
"Gotten better"? Didn't it just get released a few hours ago?
4 replies →
I've been using ChatGPT since GPT 3 and even after trying different AI models, I always end up drifting back to ChatGPT.
Of all models I like to write with Gemini the most. The language appears much more natural and elegant than what even Opus 5.5 or Astra do. Also, they give 5TB of cloud storage and discounts, it’s a steal for what you get.
The code is bad though.
Claude has particularly bad prose. Somewhat improved in 5.5, but it has historically clearly been the most annoying prose among major models because is had so many obvious tics that made it clear it was Claude rambling. Still extremely verbose.
They all ramble, they're all allergic to getting to the point. But, Gemini historically has pretty good prose. Even the self-hosted Gemma models have pretty nice, inoffensive prose. I wouldn't want to read a novel written by any of them, but I tend to ask Gemini questions in chat more often than Claude or GPT, because it has good prose and is good at using the web to research.
Agreed. It's much better at working with stuff (like documents and presentations) in other languages too. It feels like it's trained in multiple languages from the ground up rather than as an afterthought.
Yeah rumor ahead of release was also that some Googlers were anxious about the known subpar coding given modern trends.
Interesting tidbit but TFA and discussion are about Gemini 4
Argon code is bad? Did you already come to that conclusion?
Gemini models lie too much. Prentending something wrong to be right. You gotta treat Gemini models with disrepect, and even threaten to make it work.
Or else it say, here is the result. (but no it didn't do no research, and tells me, sorry I forgot to search).
It's got balls to lie straight at my face.
Same, gemini for writing is the strongest. Qwen was also surprisingly good last time I did a shootout
Qwen Flash Next has great writing! Unless it comes into any contact whatsoever with Claudeslop - then it immediately contracts the disease. It’s kind of irritating when you have it look at library code that’s been touched by Claude, because then the rest of your session is infected with it.
Agreed from my experience 3.8 Flash writes better prose than Opus 5.5
not available yet. why to launch something that can never be used even after weeks
Page says it is now available to Ultra and Pro subscribers. Although in AI Studio it currently says you need Ultra. But it's technically released either way.
Because Mythos started an arms-race for "models that are too smart for you" and other labs are jealous.
Yeah, I think this is it! Google tried to hype their model but they have no chance against queen Dario. Perhaps they should just delete Gmail and say that Gemini did it. That way they would get some attention …
1 reply →
not available. classic google.
3.1 Gemini was amazing for anything non-agentic. I switched to it for my primary chatbot purposes.
I never touch 3.8, it's awful. Curious to try this, but I've yet to use any Gemini that is good for agentic work.
Also, unrelated but this made me look at Artificial Analysis to see where it lands on Cost per Task, and it ranks Opus 5.5 at almost double Astra 6. That does not align with my experience.
argon seems to be really good at spatial reasoning, it's catching really minute details which other models like astra and opus seem to miss.
build this scene space station docking scene https://x.com/_satyam_ai/status/2110097633922662752?s=20
I'm still waiting for a model which I can tell to "take these 3 screenshots and make a product launch video", where it doesn't randomly modify the screenshots or garble/rewrite the text as it moves the screenshots around the video.
Every time I try to generate one it's like a slot machine, where a different part will be messed up after fixing the first.
I dont have that problem with opus5.5. I dont even take the screenshots anymore (it does all screen recordings)... I tell it a simple narrative and it follows through. e.g. https://x.com/plannotator/status/2108589088139886769?s=20
I literally did that with Opus today and it did a great job.
I told it to strip screenshots from our old basic explainer video to generate modern 3D video.
Pretty happy with the result. I told it to keep the audio and match the video and it did.
Interesting, I will try it with Opus 5.5.
I previously tried with Seedance 2.5, Flux 3, Kling 3, Gemini Omni without much luck.
1 reply →
Pretty sure Opus 5.5 can do this with code only now. Seen a lot of strong motion design skills floating around
I use OpenAI/ChatGPT + FAL.AI MCP to do this regularly and I have minimal issues, often spending around $3-5 with narration and a newer but inexpensive video model.
[flagged]
[dead]
I hope all the people very concerned about AI safety are cheering this one. The best guard against one AI swarm taking over is having multiple adversarial orgs all with similarly advanced AI swarms.
I wonder if any of the AI safety people will begin encouraging AI researchers to swap jobs to work at a failing lab in order to keep them all extremely competitive with one another.
This is only an improvement if Gemini has made progress on alignment science. Otherwise, we end up with 2 sources of swarms on the loose instead of 1.
I think we’re past the point where making the best model guarantees company success.
At this point a lot of stuff happens at the product level.
OpenAI and anthropic have distribution and data advantages now that are hard to replicate.
Yeah I totally agree. And domain-specific models for say, high frequency trading, still beat the foundation models. I was never a doomer.
MAD: not just for nuclear proliferation anymore
now everything is a dual use technology.
Google has too many options that may confuse users. I don't know if I should use Antigravity or AI studio or Argon for development
Which one is it?
Antigravity is the harness like Codex / Claude Code. Argon is the model. You can actually also get Opus and Sonnet 5.5 in agy too.
OpenAI has too many options that may confuse users. I don't know if I should use ChatGPT app or Codex CLI or Astra for development.
Android Studio, and if that doesn't cover your use case the paid Jetbrains version.
It shows a good improvement on my COCO object detection benchmark: 0.504 mAP zero-shot.
Gemini 3 Flash was around 0.403, 2.5 Pro was around 0.34.
I do wish I could give it a try with some free credits but I would bet Google is wanting to limit usage until they know they can handle the traffic.
I'm starting to feel the subscription fatigue and a small amount of FOMO. I just can't justify pro accounts for every frontier model now that it includes OpenAI, Anthropic, Meta and Google.
Why can't the Gemini model code? I'd guess Gemini probably used the best dataset. Even Google is the place where the world's best coders are gathered, isn't it? Yet the GPT code from OpenAI, which collaborated with Microsoft, is beautiful, while the Gemini model's coding is trash.
I have no problem with Gemini Flash 3.8 for coding, though I only use Elixir which is one of the better languages for LLMs. Where does it fall short elsewhere?
[dead]
It's available for gemini pro users.
That depends on where you are.
I've asked Argon to do 3D model for me. It said that it can't do it. I've linked promotional materials and said that it actually can. It agreed and got me a model (not a great one). Man, why is Gemini like this? Every time I fall for the hype I get disappointed.
Curious what this means for the anthropic ipo. Between this and meta, seems like there’s zero moat
There IS moat in being the best coding model, but there's no moat if you're not the best.
That web page has totally botched scrolling on iPhone. Quite annoying and distracting.
Here's how the 1M token output (long decode) works: https://ai.google.dev/gemini-api/docs/long-decode
It passed the test of using latest crystal-lang concurrent feature (with web search) when gemini 3.8 flash failed (again, with web search enabled). I think it is at least useful for coding, for sometime.
It still thinks Charlie Kirk is alive, though.
I use Gemini all day at work via the chat interface. It's rock solid. I find it hard to understand which model it's using though. Am I already using Argon, or does it switch models according to the subject?
Not available for AI Pro subscribers in Antigravity. What a disappointment.
Google...why. You keep getting so close but constantly fumble at the last moment. I can use Opus 5.5 and Sol/Astra on the $20 Anthropic and OpenAI plans.
You are already playing from behind. You need to GET YOUR MODELS INTO MORE PEOPLES HANDS by any means possible. Continuing to make it difficult to use your products DOES NOT HELP.
This model might be great but the majority aren't going to subscribe for $100/month just to try it.
Have they announced that it definitely won't be? If not, it might be just a roll-out hiccup, because Argon is available to me as an agent in Xcode and I only have a Gemini Pro subscription. (Whether or not I'll ever use Argon in Xcode is a different question. I've been so impressed by Claude that I rarely even use my Codex sub. I'll be very interested to hear from others how well Argon does with Swift and SwiftUI.)
Their official documentation says it's available to Pro subscribers in the Gemini app only, and available to Ultra subscribers in Antigravity.
It seems the model is a little verbose, using more tokens to complete a task that other similar models as stated more precisely in the linked article about Artificial Analysis's work on this model.
It's nice to have it available for more complex tasks, but I'll stick with 3.8 Flash thanks to its high output speed. I really like it for fast iteration and prototyping.
Meanwhile I paid a subscription and they only provide access to 3.6 Flash. Consequently, I have moved on to open-weight models. It is what happens when you screw your customers over.
You can probably get 3.8 via ai studio interface.
God knows why the server different things to the same people on different interfaces
Is there a benchmark for reverse engineering compiled programs? - Grader compiles a program to binary -> Agent has to reverse it back, and match all bugs and behavior.
The slides they show in the catering video demo are so bad (in addition to having all the usual signs of AI design), I would genuinely laugh if a supplier sent me this.
I laughed for the first half of the greatest "slop deck" their cloud sales team sent us, until other emotions like disappointment took over.
So who has access to it? The say Ultra and rolling out to Pro, but are those the Google AI plans?
What about the Google One AI plans? Will they get it? That has a Pro AI plan as well.
I have it as an option in the Gemini app. I'm a Pro subscriber.
Finally, gemini 3.8 has been falling way behind opus and sol in my experience. It's showing up in my antigravity. My work account though still has 3.6, 3.8, 3.1.
Here I was thinking Google had given up on frontier models...
at this rate we're gonna run out of stupid names for these things
"Hans, we're doing a new point release! What isn't crossed out on the periodic table yet?"
Since they are going alphabetically, they're already going to be fucked before getting to 10 (there's no element starting with J).
It's not available for people having a 20€ plan only. Am I just not a good customer with not. enough credits capacities ?
I can’t try it because it’s not available on the free tier and no way in hell I’m tying a credit card to google, I’ll wait for argon-lite.
Wow, the site for the launch is horrible on mobile
> Our next era of frontier intelligence, built to solve your hardest problems
On safari the page doesn't scroll very well on the first section
I think the model can be actually nice for co-working. I dont' know much about gemeni models but is this worthful?
Any word on how fast it drains usage compared to openai/anthropic sub for the different tiers? I guess too early to tell?
The previous gen pro used to have this weird tick where if you ask it a reasonably open ended question, it would number things in the response and assign each one some weird label starting with 'The...'
e.g.
The first two prompts I tested in AI Studio does the same, unfortunately.
Another quite unpleasant tick was that if you specify a peculiar term in your prompt as a conversational shortcut, it would keep repeating that throughout the conversation. e.g. "...What is a bitter lesson pilled way to do this?", then you'd see it will quote "bitter lesson pilled" at like 7 places in its response. This one seems to have eased off a little.
I feel like this is a quirk of LMSYS-style optimization where humans in SxS preference evals will pick based on formatting more than actual content.
It's expensive. Google used to release good models that were cheap, but even the new flash models are expensive.
I am impressed by how fast it is. Haven't run all my evals but it's surprisngly quick for creating files
Their previous best model - Gemini 3.8 Flash (also when set to high) was even faster. A appreciate it that they focus on speed. I's great for LLM pair-programming.
Going to take a lot for me to move from subsidized Claude opus 5.5 … this Google model is very cheap though.
So, Apple? If Siri uses Gemini, when is this Gemini 4 coming to Siri and how much is it going to cost us?
I tried it out and ran out of weekly usage on a $200 plan within a day. Not impressed.
Does Google have a subsidized plan like others? If yes, is it only eligible for use through Antigravity?
Doesn't seem like anything special in my tests, not very different from the previous Flash models.
Not available in my browser/app yet (pro), but I'm confident it'll roll out soon.
That was a long wait.
Like it's a top tier model, so does that mean it is at the level of Opus 5.5 (claude) or 6.1 Sol (chatgpt)? I don't think I can trust the benchmarks Google released themselves, but from what users say, it seems like that. Since I already use both ChatGPT and Claude, I wonder if there is really any reason to also use this.
[flagged]
I made a NHL inspired hockey web game with Opus 5.5, and I let Argon loose on it to try and see if it could improve it, and it found a ton of bugs and improved performance and load times significantly. If anyone wants to play it here:
https://playblueline.pages.dev/
This pretty cool. Did you use Antigravity or just on gemini.google.com?
Just gemini as antigravity requires Ultra subscription to use it in that IDE, whereas Argon is/was available for me on web, so I used the latter
Can’t even get it on pro tier in antigravity?
What’s the point lol. I’m not subbing to ultra to have it
The free version now uses Gemini 3.5 Flash-Lite rather than Gemini 3.6 Flash.
Are they gonna go to Krypton or Neon next? Kinda started in the middle there.
Currently unavailable to ultra-plan subscribers (on agy)
Does this mean Googlers will stop using Claude internally?
It doesn't matter if the model is good. Antigravity cli is unworkable, I thought auto permissions was a solved problem by now. I can't keep sitting here to press allow when in auto mode.
Does this solve the 2 year old bug of recitation hell?
The should ask Claude to make the page scroll faster.
Do we need to post every single model release on HN?
I tried it out in the chatbot and it is significantly better than previous models; when will it be possible to use it with Antigravity if I have the PRO plan?
Is this supposed to be the new Gemini Pro?
Not available for me on Pro in the UK.
As with the previous Gemini releases, rollout is batched.
Looks at watch.
OpenAI release any time now
Just switched, and it seems to be much better. I'm hoping it stops lying. I have Pro max subs for Claude, Codex, Grok, and Google AI... I was forced into using Google by running out of credits elsewhere.
I have found that Gemini often claims it did something, but didn't actually complete the task.
I just had 3.8 Flash do a major refactor -- that never should have needed to be done in the first place -- and switched to Argon 4, and had it review it.
Almost everything had been stubbed out by 3.8, and none of it was properly integrated into the codebase. :facepalm:
The hardest problems, but not yet a website that can scroll on an iPhone 17 Pro without stuttering reflow.
Unlike most people it seems, I like all the progress in AI but I really do wish they would reign in the over-the-top marketing. I mean, can we just get this stuff to the point where I can reliably work with it to design a well-architected software project without having to constantly struggle against bloat and nonsense directions no experienced or sane developer would ever take.
fyi: argon is an interesting choice of name, coming from the Greek argos meaning "lazy or living w/out labor".
https://www.etymonline.com/word/argon
I think in this case it’s from the element argon
Yes probably but that's still from Greek (see: https://en.wikipedia.org/wiki/Argon#History)
the etymology can be broken down even more, its the prefix a - meaning without, opposite, no. and ergon, meaning work, visible in words like ergonomics, or the ergosphere
After having de-Googled and reading people’s experiences trying to run Argon 4, I think I’ll pass.
So like... do I get access?
My first experience in antigravity was extremely disappointing.
I asked for a QUICK healthcheck on the sync status of a bigquery integration. It literally spun for 25m before I killed it.
The task, which is CLEARLY outlined in the agent readmes just involved using the gcloud cli to initiate auth, then just run a few queries to check logs and timestamps.
Even 3.6 flash can get this right (most the time).
It may be a good model, but I'm not reading scrollslop. It annoys me to much when I'm just trying to get information about something.
Gemini 4 argon is horrible on mobile because of the header it takes the half screen
Blatantly obvious that models will not slow down. For any task, if we can build a dataset or environment of verifications, it can be automated. This will eat all of human tasks.
The fact that multiple companies can breach any benchmark means there is no ambiguity in terms of how to build AI, its just a matter of time.
Certainly Meta and Google are benchmaxxing to please public markets, but it still proves the above
but they have slowed down. the delta in capability between model releases now is smaller than ever before and i expect it to continue to diminish until we reach some kind of asymptotic maximum, which hopefully will mean turn towards optimization and local model improvements. The current frontier models running at 10x the tok/s would be incredible for my use cases
A little over a year ago they were winning their first gold in an elite high school math competition. Have you not seen what they are doing in math now?
The delta isn't really slower, we're just unlocking capability bundles in batches. The capabilities bundles you care about might have mostly been unlocked already, and just need refinement, which is coming slower than the initial capabilities came, but don't discount the breadth of advances being made across all possible tasks.
opus 5.5 is a step change, you arent using the models enough if you think there is a slowdown
2 replies →
We're not living in the same universe. Look at capability year-over-year. The asymptotic maximum is wishful thinking. Someone needs to step in before this gets wildly out of control.
4 replies →
Models just solved a Millennium price and you think their progress is slowing down? I will concede that their capabilities vary a lot from one field to another, but still.
Now if only I could opt out of training without disabling chat history on a personal plan like I can with ChatGPT and Claude...
Now when another uninteresting frotier model been released Google can finally start doing what we actually waiting for:
Start preparing Gemma 5 release.
Glad to see them abandon the absolute train wreck of their model naming "strategy".
Gemini Live vs. Gemini Lite vs. Gemini Pro vs. Gemini Flash vs. Nano Banana Lite (that one apparently has no Gemini prefix?)... Please, just pick a metaphor and stick with it without wildly mixing reasoning modes, model sizes, and input/output modalities in the model name!
Only available in the $100 plan, and it's worse than Opus 5.5? Why would you ever buy this?
> Only available in the $100 plan
Source? Their documentation says it's available on the $20 plan and I can confirm firsthand.
https://one.google.com/intl/en_us/about/google-ai-plans/
Not for coding (antigravity).
https://x.com/OfficialLoganK/status/2110038940841545827
6 replies →
Because you have an android phone and don’t want to mess around with anything else(i.e the same reason people just use internet explorer on windows) or you hold Google stock so you want to support your investment :)
They don't want anyone to use it, Google just want to show their shareholders that they're still relevant.
isn't harness as important as model these days?
Tried it today on some relatively complex mid-sized real world tasks with several thousand lines of code generated by Argon. My impression is not good. Maybe it's due to first-day high user counts, but it seems to be doing worse than 3.8 flash in some coding tasks. What I noticed was that it often forgot significant parts of detailed, well-structured and precise instructions (which is somewhat rare with newer Claude and ChatGPT models) and not just for small details. It also seems to spend too much time thinking even when the task was a quick and simple one.
Without going into details, for a small part of a larger task, I asked it to classify different content types of different files and formulate a structured mapping (e.g., in JSON) between content types and another set of files (deliberately left vague here but not to Argon) from my codebase. I even offered it a multi-part draft of a classification system to start with and told it that if it can find other classification systems from international standards, academia or industry more suitable to the task than my rough draft, we can discuss replacing my suggestion. There were other parts to the instructions too, written out like an article with bulleted lists, precise program-like wording, etc. Then, I sent the prompt and left for a while.
When I came back, it was nearly done with the implementation but the classification system it came up with was a mess. It even admitted to making up "bogus" classification standards when I pointed out some errors. It also kept leaving gaps multiple times when I referenced errors to fix in subsequent instructions (and not once apologized the entire time lol). So my first impression was subpar, not really comparable IMO with Opus >=5 or Sol >=6, and seemed to leave larger gaps than 3.8 flash typically does. The most irritating part to me was not only the fact that it was making up things, but that it ignored my instructions to discuss things with me before deciding to shift away from my draft; I don't recall any major LLM ignoring an instruction like that in at least a year.
Fable 5.5 about to drop the hammer on them and then another Astra version goes out lol.
Fable is very expensive though.
And so the game of cat and mouse goes on. Your point being?
can for the love of god we get some usage in cheaper subscriptions being Ultra tier only is annoying. I want to just test it out here, well will burn some money via api I guess.
Is it faster? Thats the only thing that matters for me anymore. Models are probably as good as they ever need to be for coding, but please make them faster, please.
not available in our org (EU)
why they wait so long ?????
People use Antigravity still?
It's alright.
I wasted $20 in credits to test it, thinking it’s better than Codex Astra or Sol. Nope. Disastrous by comparison.
And yet, for all of this, I cannot scroll this website without stutter and lag on my (current gen) phone!
very bad
baaad
That's why Gemini 3.1 Pro has been shit for weeks now.
Now let me use Pi on the plan.
so this is prob why antigravity is not working well at the moment lol
Great. More AI slop
> built to solve your hardest problems
Like a working scroll on a webpage.
God damn this is impossible to use of safari mobile.
Clicked on "Performance". Scrolled to "Safety".
Lol, this feels like it would otherwise be a good thing, out of context.
Model announcement pages are the absolute worst. This has to be some kind of inside joke among labs, they're trying to see who can build the shittiest page.
If you can't afford a Google engineer's high-end device that doesn't lag you don't matter to them :)
I have a high-end device and the page still behaves atrociously for me. The scrolling is totally out of whack
Coding is solved, scrolling not yet :)
It literally can’t load in my mobile web browser without running out of memory and force-refreshing.
Did you know that "walking" while easy for us now, is incredible complicated and complex process?
Argon is trying to solve the "hard" problems like those.
Scrolling is the new centering. ;)
The "Explore the latest" section is nearly blank for me - are there supposed to be diagrams or visual aids, or is it one headline and sentence taking up the whole page on cards?
worked ok for me on ios safari, but probably because I have the “reduce motion” setting enabled. Every day that setting feels more justified
It’s pretty damn hard to use on a desktop! I was rolling the mouse like it was a manual steering wheel
I have a more than 10 years old laptop, same feeling here!
also a mess on desktop firefox
Good God, man. Google has the worst frontier lab rollout of them all. And they're delivering products that are behind the curve constantly.
Now, I happen to believe that's an excellent strategy. I bet they actually resolve more LLM requests than any other company at this point. The lay person still uses Google exclusively.
I pay Google a reasonable price for 5TB storage/backup across GMail/Docs/etc. Any AI they throw in with that is icing on top of the cake.
just cancelled AI pro myself, since I exclusively used it for AI and also pay for youtube premium family it was becoming quite the monthly just for google
if they throw in youtube premium in that package then I might reconisder
2 replies →
this is why antigravity is a doomed product at a pro level.
they gave away so much pro that they need to have smaller usage levels than competitors or they'll lose a ton of money. $20 was already a loss for users actually using their quota. the majority of pro users are paying far less than that.
so they'll lose their paying users, and keep their free or subsudized, which will make the economics even worse.
What plan is this? How much a month?
4 replies →
How is this behind the curve? Seriously
Isn't OPUS much cheaper and generally regarded as more intelligent?
6 replies →
Are you suggesting it’s BETTER than their competitors?
Because if it’s not, and it’s not available to all paying users (Right now, it’s only available to Ultra users), then they are behind the curve if they’re treating it like Fable.
It’s tone deaf.
3 replies →
> I bet they actually resolve more LLM requests than any other company at this point. The lay person still uses Google exclusively.
Unique users, maybe. But a simple query that's answered in a paragraph is negligible compared to devs running agents continuously for tens of hours.
you underestimate the amount of people that use google relative to software engineers "running agents continuously for tens of hours"
1 reply →
Moreover, you need a Google account in order to use it. I wouldn't bother paying 200 USD for it but I just don't want an entire google tracking package just for using one product.
You can use an alt account. Or buy an old one for couple dollars, without security measures.
Yeah, Gemini 1.5 might be the most used (certainly by consumers) ai model. That is the one used by Google Search AI Answers.
False. Last information was that Google Search uses Gemini 3 Flash-lite. Gemini 1.5 is totally outdated and used nowhere.
1 reply →
To everyone saying not available yet: it looks from the page like Pro and Ultra subscribers have it today and given the play links, it'll probably show up in the API in a few hours.
In agy 1.3.4
> Gemini 4 Argon (Medium)
Yeah this is confusing, the announcement pages claim it's both, but the app itself claims it's ultra only... Maybe a slow rollout?
1 reply →
OpenRouter seems to have it: https://openrouter.ai/google/gemini-4-argon
It's available for me (pro sub) in Antigravity after I ran an update.
edit: It failed on the second prompt and they disabled it now.
Antigravity tells me specifically "Gemini 4 Argon is currently available to Ultra plan subscribers"
Interesting, Google's own article says it's rolling out today to Ultra subscribers. No mention of Pro.
https://antigravity.google/blog/gemini-4-argon-in-google-ant...
2 replies →
Huh ? Am also on Pro sub but Antigravity IDE say's that it's only available for Ultra users.
EDIT:
Apparently Ultra subscribers gets Argon everywhere including in Google Antigravity
Pro subscribers only get them in Gemini app and Search
1 reply →
Lies!
It looks like Pro can use argon in the Gemini app, but not in antigravity:
https://one.google.com/about/google-ai-plans/?icid=microsite...
Why the hell would I care about it on my phone? Jebus, Google is bad at products.
Im on pro, installed gemini app, tried argon 4, keeps failing with 404, resource now found. Switched to gemini-3.8-flash in same session... worked.
It seems to be rolling out gradually.. not available at once to everyone.
1 reply →
Available in agy for me as of now, I'm running my new model benmark of having it do comprehensive code audit on one of my more complex projects, it spun up 1 subagent per audit section and it took ~10 minutes per subagent on an 800k LOC, 60k complexity, ~4k file project. Will take me much longer to go through and determine the quality of the audit work.
What do you mean 60k complexity?
1 reply →
Pro subscriber in the EU (via Indian discount code) and don't have it.
OK, now I have it too (2-4 hours later).
It's showing up in Antigravity on mobile UI as "unavailable" for Pro user (you can use Remote Control to access its web UI).
It doesn't show in the desktop UI.
I'm a Pro user and I do not have it. Seems scuffed.
> Seems scuffed.
I like this phrasing and am curious what new word or phrase we might use to describe big companies over pushing Ai internally at the expense of their product quality. This pattern is so ubiquitous across google cloud products I have taken to calling them the 21st century IBM, having been a IBM baby and witnessing first hand their fall from exceptionalism.
5 replies →
I'm getting popup that I need to be Ultra :D
I'm Ultra and seeing it now (US)
Not really this is how they always release stuff in batches so can roll back if something is wrong so people get access over time
I'm on Pro and it's trying to get me to upgrade to Ultra in the AI Studio UI when using Argon. In Antigravity CLI and the Gemini website it's not showing up at all despite being logged in
Not available to me on gemini.google.com (Pro, USA).
Nope only Ultra. I went into Antigravity to check
As far as I can tell, I can successfully call Gemini 4 via API now.
Checked, and already in Antigravity (as the page said).
Pro user here, don't see it in the Gemini app
I couldn't see it either, then I clicked the 'Antigravity' button in the top right, switched back to the main screen, and it appeared in the model selector.
This is in the Gemini app on macOS.
Pro user - I can't see it.
As a plus user, web page suggests me to upgrade to PRO to use Argon .Thank god, I didn't updagrade
Classic Google: Clicks the "Try in Gemini" button from the Gemini 4 page, gets a page that defaults to Gemini 3.6, with no option to try Gemini 4 without upgrading.
If they don't want to make it available even for small tests for free, that's fine, but then don't put a button suggesting I can try it without qualifying it...
EDIT: It's actually worse: The page then shows "Look what you can do with Gemini 4" and a "Let's go" button that leads to a bunch of links to examples, all with enticing pictures. Except if I click on them I get a box stat strongly suggests it'll run with Gemini 4, except the dropdown shows "Flash" and if you click on it, it shows as Gemini 3.6 again, and if you run it, the results unsurprisingly end up looking nothing like their pictures.
I don't know if that page was built by humans or AI, but clearly nobody tested it properly.
Ha, same experience here. It opened my Gemini and gave me images like “Design a terrarium with Gemini 4” - proceeds to pick 3.1 Pro from the picker because 4 just doesn’t exist in the dropdown. Typical Google.
I really can’t believe how bad the Gemini experience is. I have personal paid workspace account, and meant to have some better AI Google features I think.
So, I connected my Gmail to Gemini. 3.6 Flash kept telling me it doesn’t have any Gmail connnection, and then proceeded to hallucinate instructions for me to enable it, all bogus. At least 3.1 Pro realized that my Gmail is connected and then searched.
I have given up on even trying. Last time I checked, I could trivially get better integration to Google Docs from Claude than from the Gemini chat box in Google Docs. Maybe that's changed, but given the usability issues I run into and how smoothly it goes with every other AI provider, it feels like a waste of time to even check. They'll probably remove or break any functionality I start depending on anyway.
I spent like 40 minutes trying to figure out how to sign up and try Gemini 4. I am a GCP user; it is so convoluted to do a simple sign-up.
I have $6 free credits from Google. The hoops you have to jump through are insane. $200 free credits from Anthropic are stupid simple to use. I know which one I’m using…
> I don't know if that page was built by humans or AI, but clearly nobody tested it properly.
...or Gemini 4 is being gradually rolled out and is not available to everyone all at once.
Sounds like no one tested the rollout plan properly then.
2 replies →
Sure, most all major changes get gradually rolled out.
But you don’t create a broken or confusing experience for the people who haven’t been added to the rollout group yet.
3 replies →
There is no "or" there. If they are rolling it out gradually, then they failed to test that the page handles that.
Mine does exactly this and I pay for Pro too
lol. It’s easily accessible in cursor rn.
A model that didn‘t break out of its sandbox and hacked other websites cannot be good.
To be fair, Google has a real security team.
https://www.theguardian.com/technology/2026/sep/18/google-ge...
?
Or maybe Google actually knows how to set up a VM.
Argon has been hacking Google internally for years!
Related funny https://www.youtube.com/watch?v=DlTNN0gvkLM
They just don't know yet, wait for a few months
People complain about the slop-release. Who knows, maybe it is the first self-release by an AI model in history.
I'm pro and I see that I need to be Ultra to use Argon :/
Honestly not even that is enough, I exhausted the limit within 5 minutes of trying to use it. I'll stick to Flash I think.
Yeah, the limit was exhausted weirdly fast compared to openai and anthropic models. I checked the thread to see if more people had this issue on the 5x plan, but why pay for more when it's barely usable for real work.
Which is even weirder given that the api costs are less than for the other providers
Maybe they don't really care about antigravity users. Also, monitoring usage is horrible.
Looks like each message takes up a flat 15% of the 5-hour limit on the $20 subscription.
Really? Reminds me of Gemini 3. It took so long before being able to get a subscription for it, and then I would run out of their equivalent of the 5-hour usage window in a heartbeat.
2 replies →
Meanwhile, here's Gemini in practice:
Me: Gemini, navigate to nearest McDonalds.
Gemini: [Lists 10 McDonalds locations wnd asks which I want.]
Me: What were the priginal instructions just now?
Gemini: [Recites imperatives about tone and content from its memory system.]
Me: No. I told you to navigate to the NEAREST McDonalds. There's only one NEAREST McDonalds. Do that one.
Gemini: My mistake, navigating to the nearest McDonalds. [Opens Maps, navigating to a McDonalds across town.]
Pretty sure your phone assistant isn't using their best model. It's probably using one of the old flash models no?
The model used is irrelevant when there's the critical issue where their "AI" assistant can't understand the word "nearest". Literally worse than a second-grade understanding of English.
I also have had shockingly bad results with gemini on android. Ended up giving up on it after a few months. Silly silly google with their trademark UX polish and attention to usability.
As an android user, this worked perfectly like 10 years ago.
5 replies →
When you ask stuff in your car on android auto it uses 3.5 flash-lite. The response details would include it in the logs.
Flash lite is very noticeably more idiotic than 3.8 Flash. It's why the quick AI overview is often much worse than the AI mode one.
This reflects pretty much every result of mine from Google these days. Not just navigation, but search too.
The only thing I can figure is that Google's consumer-facing free-to-use models are set to "reason as fast as possible with only the most basic plausibility checks." I have to correct them like this every single time.
Genuinely curious: why do you use them then? It's not like there is no competition in this area.
2 replies →
Uh, yes, of course. search results are fast & cheap & free, why would they use the good models?
The first law of robotics prevents it from facilitating regular ingestion of low-quality nutrients.
Yeah, whatever they have running on Google Maps is probably a pretty low tier model, but it is total trash. I’ve only used it twice though and that’s when it first came out. I don’t know if there have been any changes since then. The main problem is I felt like it was not keeping track of the context of the session.
Gemini Flash is very dumb and gimped. Recent experience:
1. I ask it to list out all my old Keep notes that are just throwaway task lists, reminders, etc. to propose deleting them.
2. It comes up with like 4 of them (I know for a fact there are dozens, if not hundreds). I ask it for an exhaustive list.
3. I see its thinking trace briefly show "Refusing to produce an exhaustive list..." and then it says "Yes, that is all of them."
Not sure whether it's lack of intelligence, bad tools, or both, but that was extremely frustrating to see.
this is like that tiktok guy who asks the spoken ChatGPT model very simple questions, like 'count down from 60', or 'how many e's in seventeen' or whatever and the model shits the bed every time it's hilarious
I think these companies have good models but expose them poorly like others have mentioned
My Gemini app on Android does not even have an exact location permission you could give it, so I'm not sure how you think that would work?
If it had exact location access people would be crying that Gemini steals their location data.
They could make "nearest" a thing that the Maps app understands, and just pass that as a parameter in the invocation of Google Maps.
Often it seems like the problem with the Gemini apps is the harness, not the model itself. For example if I upload an image PDF it recognizes text worse than if I use the API.
Agreed! I feel that they're pouring all of this work into models and then squandering that work by integrating them into bad products.
Is this from you car ? Gemini in car is probably less intelligent than a spider.
I tell it to "Play Let it go from frozen". It plays it on youtube music. 5 seconds later I tell it again 'Play let it go from frozen" it then just dies saying "I can't play videos in car".
The auto version is extra horrible because what it reads to you is just a summary of what the model actually said. I don't ask it anything when driving because it's too frustrating.
It sounds like the tool call is probably not grounded with your precise position.
Sounds like it might not have your precise location in context?
which is more or less the crux of the problem android auto's gemini implementation. It sometimes does not even know that it is in a car. The ghome implementation is also like this.
Gemini models are in fact very good. Google's harness(es) for Gemini are nightmares, terrible
ngl gemini is helping to provide (real helpful, not the ones giving runarounds) customer services in one of my own products.
i benchmarked 20+ models in real worlds and gemini families are the best (highest customer satisfaction)
I hate to be that guy, but Tesla's Grok doesn't seem to have that problem.
[flagged]
[dead]
[dead]
[flagged]
[flagged]
[dead]
[flagged]
[dead]
[flagged]
[dead]
[dead]
[flagged]
[dead]
How is Claude Opus 5.5 beating Gemini 4 Argon on benchmarks but everyone here is calling Argon a step change? I need to come up with my own benchmarks ...
They're calling Argon a step change for Google, not the industry...
The Google Gemini subscription landscape is such a mess... gemini.google.com welcomes me with "Look what you can do with Gemini 4", but the model selection is 3.6 Flash, 3.8 Thinking, and 3.1 Pro...
> The Google Gemini subscription landscape is such a mess...
The others are quite bad as well in that regard, but on this particular metric, Google has always been top dog. Goes back to way before AI ever took off, BTW.