Comment by postalcoder
12 hours ago
I wonder how big the Pro model is that Google is using behind the scenes to train these smaller ones.
Going on baseless speculation, the lack of accompanying pro models with these flash releases either means: 1) the model is too big to be economical, 2) google doesn't have the compute to serve the big model, 3) their big model has too many alignment issues to serve to the public.
edit: looks like benchmarks are up on https://artificialanalysis.ai/models/gemini-3-6-flash. It's solidly middle-of-pack. However, if you want to be most fair to flash, look at the intelligence vs time per task and intelligence vs outputspeed benchmarks. This is a very fast model.
edit 2: I use antigravity from time to time and in my experience, 3.5 flash is an underrated model, so long as you know what it's good for. It's very good at frontend (much better than gpt 5.5) and it's fast, so it's a great tool for iteration. I expect 3.6 to be no different.
It's also very possible that they know their big model underperforms chatgpt 5.6 and fable by too much, so they are focusing on what they can get wins in like speed instead.
That and/or the business case isn’t as clear when serving enormous models? You’re constantly stuck in a red queen’s race where your profitability window is increasingly measured in weeks because the Chinese are right behind you.
For small models (which are probably distilled from their big ones) you can serve them economically all the time and not hemorrhage money.
> For small models (which are probably distilled from their big ones) you can serve them economically all the time and not hemorrhage money.
For smaller models, you're competing with DeepSeek V4 Flash. (Which I think is a 284B A13B?) Subjectively, this feels about as smart as Sonnet 4.5, give or take. And it costs $0.09/$0.18 on Open Router, compared to $1/$5 for the latest Claude Haiku. See https://openrouter.ai/deepseek/deepseek-v4-flash#providers The developer antirez of Redis fame uses this as a local coding model.
DeepSeek did some extremely clever research on hybrid attention to get the prices that low, reducing per-user context cache sizes dramatically.
So, no, when it comes to low-price models, the US models probably can't sustain their current margins there, either.
The Chinese have been right behind OpenAI and Anthropic for ~18 months now.
DeepSeek didn't do to OpenAI and Anthropic what nearly everybody claimed they would.
Every single person on HN that loudly proclaimed the end was nigh for GPT & Co. due to DeepSeek, was wrong. They were humiliatingly wrong, and they'll never own up to it. The reason those people were so very wrong, is the same exact reason the Kimi crowd is wrong now. And it's very obvious that they're wrong, but they have intense emotional blinders on. Their thinking process is hyper emotionalism: they want a certain outcome, regardless of if reality aligns to that or not. They're making emotional wishes about how they want things to turn out, and pretending those magic wishes are grounded in reason.
It takes enormous resources to run something equivalent to GPT 5.6 or Fable. Nobody can or wants to do that outside of very limited situations - if you can just reasonably pay as you go instead. As it turns out, you can just pay as you go with GPT and Fable. Their businesses have gotten radically larger since DeepSeek launched. Get it yet?
Domestic China is the only very large audience for their own models, so long as OpenAI and Anthropic stay top tier.
All the hype online from the forums about Kimi, is worthless: those people hyping it can't even come close to running it locally, which is the fantasy. So why are they hyping it? Why did they hype DeepSeek just the same, and learn nothing from its total failure to actually take down OpenAI and Anthropic? Rhetorical questions with obvious answers.
Kimi poses zero actual threat to OpenAI and Anthropic. Those companies will continue to pile up the subscriptions and API usage. Check out GPT's subscriber base today vs when DeepSeek launched. Get it yet? When Model X launches out of China in a year, we'll have this same conversations all over again, and the hypsters will have learned nothing.
While the Kimi fawning is endless, OpenAI will just keep piling up subscriber counts, and Anthropic will keep piling up API usage. Then OpenAI is going to staple a gigantic ad system onto GPT. China can't compete in the model-as-a-service business globally, for the exact same reason they failed so miserably to compete in search globally.
12 replies →
That's the only explanation that makes sense. If it was frontier but cost or compute were limiting factors, they'd release it at an obscene price for the bragging rights. Google doesn't care that much about alignment, and I don't think it's likely to be significantly different than 3.5 anyway. The only reason it would need to be soft-canceled is if it's terrible, and has to end up in a ditch like Llama 4 to avoid shareholder panic.
3.5 pro was clearly a miss. It should have been in prod mid may, not MIA in late July. The brain drain at deep mind is a clear indicator that the people who know the most think that they can’t stay at the frontier.
Antigravity NEEDED to be game-changing. Without the stream of data that Claude, Codex, and Cursor enjoy there is little chance of getting an effective reinforcement learning loop. For the first time in its history, GOOG is at a meaningful data disadvantage, and apparently a cultural one as well.
2 replies →
It's also what Pichai literally said in a recent interview, that Google is not doing well in coding and agentic tasks.
https://www.searchenginejournal.com/pichai-says-google-is-a-... (link to the actual podcast interview source within, this has a summary)
“Google doesn't care that much about alignment”
I don’t think this is necessarily true, did we all forget how much Google cared about alignment that their AI wasn’t able to render a white polar bear?
Yes agreed - I wrote this up a while back https://martinalderson.com/posts/whats-going-on-with-gemini/
My view then was they are optimising the models for inference ability on their own hardware AND use cases, which is often speed and time to first token.
They've somehow seemed to end up with terrible compute shortages, which again is surprising given how good Google is at infra deployments AND have their own hardware. From rumors out there they are turning down enterprise deals for Gemini because they don't have the compute.
The problem is they're falling further and further behind on frontier class on coding especially, and since I wrote that article it's got even worse with open weights models undercutting them on price AND intelligence.
They include an LLM response with every single Google search, whether it is warranted or not. That scale is, my guess, many orders of magnitude higher than what OpenAI and Anthropic serve. And for Google none of these are paid interactions since their LLMs do not (YET) insert ads into the responses.
So my guess is that Google will continue having compute shortages until the Gemini enshittification starts.
1 reply →
There was some recent reporting that a July release of the Pro model got pushed back for exactly that reason. Its performance was not good compared to the OpenAI/Anthropic big models. They are having a lot of problems with posttrain.
This is the feeling i get too. Cant produce quality, but can produce something that is super fast...so take the wins where they are.
We don't have enough fast models, so I see this as a positive. I just test drove Gemini Flash Lite and it's crazy fast.
1 reply →
For a coding LLM specifically, when is fast a good tradeoff for quality?
3 replies →
I personally doubt that.
It would be a shame if they cannot beat Kimi K3 or Qwen3.8 Max, both of which are claimed to be Fable-like. If that is true, it will be [or would be] the first time a major American lab falls behind a Chinese competitor.
Google can't compete with China, neither can Meta. Only two labs in the US can keep chucking billions at the frontier race. Everyone else has a real business to run.
China can keep up because it's cheaper to run a frontier lab there. They also have more researchers and a stronger cultural inclination for this sort of thing. And I guess the business case in China doesn't have to work as well as it does in the US.
1 reply →
> focusing on what they can get wins in like speed instead
Speed as a differentiator has always been Google's thing. They (used to?) show the microseconds it took to query & rank web-scale search results. Chrome, notoriously, focused on speed at the expense of resource use. The very many efforts to efficiently speed up Android & its runtime since its inception, and so on...
> their big model underperforms chatgpt 5.6
Possible but TFA claims:
That sonds like they can't compete with 3.5 or 3.6 so they must increase the model size and are training v4.
Didn't they already acknowledge this?
Paywalled article, but the headline is basically all you need: https://www.bloomberg.com/news/articles/2026-07-16/google-ge...
I choose fourth option.
4) googles big model just performs worse than K3 and GLM so they choose not to embarass themself.
Like I love Gemini and use it a lot to one-shot whole MR with huge contexts, but its just much worse when its come to tool use and agentic coding.
or they just don't want compete in coding space ???
they have search,youtube,android,office suite like gmail,maps,spreadsheet etc
coding is the least of their problem/priority
I wonder if the broad use of AI overviews on Google search results is having an impact. Maybe the numbers make it more profitable to use their compute on several billion searches a day rather than selling API access.
I think it's a safe bet that Google seems more interested in making a model that improves Google rather than making a model that improves workers.
Fast, light weight, ok intelligence. Perfect for serving 20B+ prompts per day mostly surrounding banal human things.
OAI and Anthropic's cloud spend can cover the revenue gap, as Google is already capturing a large chunk of those guy's revenue.
Not to mention internal use cases, such as prediction-related tasks like serving ads.
The AI mode on Google search is pretty impressive. Helped me figure out what a bunch of stuff I was seeing out the window was while traveling.
Microsoft also seems to be working in this space. They recently released this:
https://huggingface.co/microsoft/bitnet-embedding-0.6b
It’s a small multilingual embedding model designed for things like search, RAG, and semantic similarity. It supports a fairly large context window and is designed to run efficiently on a CPU in a GPU starved world.
The interesting part is that it builds on BitNet, using ternary weights of -1, 0, and 1 instead of the usual floating-point weights. That should make indexing and searching large amounts of text much cheaper without giving up too much accuracy.
AI overview is just a summarization of the top 2-3 results. Of course at Google scale that will still need a ton of compute, but the requirement for generating an overview is many orders of magnitude lower than asking the same question in Gemini.
based off what?
1 reply →
Artificial analysis always seemed sketchy as hell. If you read some of there methodology you’ll see a lot of <=3 repetitions on a particular pass for a given model. So low for calling a frontier model over the public internet ????
Flash versions were often ultra competitive and their best in the range along with openai mini models. Always been gemini most useable and best model with nano b. Frontier is much more competitive. Anthropic haiku is like 2025 flash…
It seems like there are some credible rumors that Google is actually winning in terms of actually building models that work and don't lose money- between how they're able to price them, the TPU advantage and their capex advantage (being able to raise debt + just having a lot of cash - well I said not lose money... more like not go bankrupt).
From the outside they look like they're behind in terms of frontier models, but I think they might be the best positioned to not go out of business when the bubble pops.
Also look at the fact that they've been able to deploy AI-assisted search at google scale. It must be another order of magnitude larger (at least) than the model deployments for OpenAI and Anthropic.
Of course unless you're inside Google it's impossible to know for sure.
In terms of open models, Gemma 4 beats the pants off everything else to the point that paying for APIs becomes hard to justify. Qwen has the meme-share for coding, but it feels much less well rounded. I have no doubt that Google have both the infrastructure and the expertise to curb stomp everyone else, should they resolve in earnest to do so.
Lest we forget, "Attention is All You Need" came from Google.
> "Attention is All You Need" came from Google
It also came directly from the university of Toronto, and the university of Toronto seeded all American frontier labs (including Grok (why do you think they could start so fast))
Interesting, glad to hear. We have gemma4 at work, and I was considering localhosting qwen, but gemma4 is so far behind the Opus and Fable I have at home that I've decided to hold off for another model release.
1 reply →
Are you suggesting Gemma beats GLM 5.2?
1 reply →
How long until Gemma 5 hits?
It's rumored that Gemini 3.5 flash has a >50% margin, and I'd imagine 3.6 flash is even higher.
I do not think OpenAI or Anthropic are actively chasing margins - though, Anthropic is supposed to be profitable on some form of non-GAAP accounting...
I suspect Google isn't really interested in seeing how far it can get dragged into a race of selling dollars for $0.25, and is more interested to see if it can stay in the race selling $0.50 for a dollar - when everyone else is losing or barely breaking even.
It kind of doesn't make sense though, because typically a large org like Google can afford to crush competitors on pricing. They could probably even go toe to toe with chinese model pricing for years without feeling it.
Maybe they don't want to price war with the other labs so they can comfortably maintain healthy margins on selling them compute?
2 replies →
google has to make money. flash is awesome. you can run it free on their infra and the performance and latency is excellent for what you wait and pay for right now, with great perf per watt. every person in the world going to google.com runs it. every query. its far larger than free gpt, localhost qween and what not.
it's their pro that isn't awesome at all. in fact, their pro kinda suck now that everyone else woke up.
That "TPU advantage" might be slowing Google down (though likely not as much as their internal bureaucracy).
Porting CUDA-based research, debugging, and overall experimentation speed is likely slower.
The GPU is still king for training.
But maybe the TPU advantage is in inference? That's what I assume because the number of compute cycles are going to be all in inference vs training. So they could train on GPUs if they want.
lmao, you know all Anthropic models are trained on TPU right?
1 reply →
They basically don't exist in the currently most profitable LLM market (coding).
Yes, subs like codex are heavily subsidized. But API billing has massive margins and that's what enterprises pay.
On the other hand, the consumer side of the market seems to be less competitive right now.
OpenAI's new Mac app doesn't even have a normal "Chat" option now. OpenAI might be chasing coding and b2b sales more now that they realise very few regular consumers pay for subscriptions.
Does it have "massive" margins? Afaik no one has said publicly what margins there are on an API call?
3 replies →
Logan Kilpatrick said on an interview not too long ago that flash 3 and 3.5 are the same pre-train. all gains on top of 3 flash are post-training
Maybe, but they said they have “started” the Gemini 4 pretrain. So not having done any significant pretrain in a year or so seems odd to me.
Pre-trains take a huge chunk of your compute offline, incurring both an raw expense (24/7 max power for all training clusters) and an opportunity cost (could have sold excess compute during that time). They also don't come with any great guarantees, as lots of techniques look good on small scale and crumble or plateau once scaled.
Friend works for Google vendor who generates data for training. His team alone is 200 people (in US).
He says there are many similar vendors and teams with thousands of people in India and other countries.
More likely they don't manage to advance benchmarks on the SOTA level anymore. In other words: They can't beat 5.6 nor Fable
I think it's 2. I frequently get told there's no capacity for Pro and the query is answered by Flash with extended thinking. And tbh it's hard to tell the difference between the two, especially if you're not coding with it.
It's hard to tell the difference because they nerfed Pro to oblivion, it used to be much, much better model (even for non-coding/chat)
Maybe it's like Meta not releasing the big version of Llama 4 a year or two ago
2.5 flash was absurdly capable on a cost basis
I wonder if they waited for the new TPU generation to train a larger base model.
"3.5 pro is testing with partners! will hopefully land soon."
https://x.com/OfficialLoganK/status/2079596415509303596
> the lack of accompanying pro models with these flash releases either means:
Rumors say 4) it didn't perform well, especially in coding so has been delayed
Or perhaps 4) it's outcompeted severely by other models & releasing it would only tarnish their name