Comment by wxw
3 days ago
They need to release benchmarks against Luna/Terra. Luna is much cheaper which feels like it undercuts the need for Flash.
I've always considered the Flash series of models to be for low-cost, high-volume, mostly text-based use cases (e.g. summarization, parsing, formatting), emphasis on low-cost.
[edit: ah, benchmarks here: https://blog.google/innovation-and-ai/models-and-research/ge...
more of a Terra than Luna competitor which is an interesting positioning. I feel like differentiation at the mid-tier of models is pretty difficult.]
flash-lite is more of their luna tier competitor but even still not quite there yet, but gemini's dominance on multimodal and image understanding i think really gets downplayed on this site when most people think the only think you can do with LLMs is write code
Ultimately it would track that in the real world, people will want to point cameras at things and get answers.
I pay for ChatGPT and Gemini, and while Sol is a total beast with anything text, it still poisoned my cucumber bed. Which I will be bitter about for at least a few years while the bed recovers. Gemini (even flash) is exceptionally talented at viewing photos and telling you what to do/what it is (and telling me I just misidentified the problem with my cucumbers and spraying off the "bugs" actually just spread the bacteria everywhere.)
> it still poisoned my cucumber bed
I also had really poor results with all major models from plant identification to plant treatment which is weird considering how much training material is out there.
Reminds me of recent headline: Chinese farmer kills 25 acres of crops due to LLM pesticide recipe: https://www.tomshardware.com/tech-industry/artificial-intell...
Tell me more about your cucumber bed. We started some raised veggie beds this year and my wife is relying very heavily in ChatGPT and Claude for advice on how to deal with issues.
1 reply →
That's been my association as well. I see Flash get brought up a lot in relation to things like OCR and PDF processing frequently, and a lot of other routine multimodal workloads.
Flash models and Gemini make more sense when you consider Gemini Enterprise and Workspace. Oh HN we generally care a lot about writing software. However, until Fable, Gemini 3.1 Pro was my default for doing any sort of discussion outside of software engineering. Fable is now on par with things like modifying cars, etc. But I am guessing Fable is a /lot/ more expensive to use.
And in a typical enterprise environment dealing with documents, images, and broader business reasoning skills matter. Agents are not just for code and text :)
Yes this is my impression as well. To be fair I didn't compare to Luna yet, but Gemini 3.5 Lite is a very good and cheap multi-modal data extraction model.
If anyone knows of a cheaper vision llm with the same accuracy I would love to switch
They compared against 5.6-terra on the model card: https://deepmind.google/models/model-cards/gemini-3-7-flash/
gemini flash is probably the best model for visual tasks right now. they also make it really easy to ingest videos
Crazy it's still the only video understanding endpoint. It's what I use it for and no other model even offers a competitor.
You probably can't build such a model without unlimited access to YouTube and Google has been tightening the screws on that over the years pretty systematically.
to be fair, all it's doing is sampling the frames and maybe doing transcription, if I'm not mistaken. So you can do it with the other models too, you just need to sample the frames yourself and do the transcript yourself
1 reply →
Yes, was going to say I use it exclusively for video and audio. The ability to give it a YouTube link through the API and ask questions about it is awesome
Also best at OpenSCAD, seemingly for the same reason, at least in terms of "iterate on a design, comparing visual output to target".
I've found Sol excellent at OpenSCAD.
But I don't do "compare visual output to target" as much as 3D reasoning type tasks (eg: "Build a G1 curve where the -X face meets the +Z face" etc)
2 replies →
+1 gemini models where really the only ones fullt grasping spatial reasoning even compared to opus (at least when i last cared to check it)
Are you manually rendering previews of its OpenSCAD output to create images for it to review, or do you have a workflow that automates that?
2 replies →
Ah, multimodal is a great point. I'll need to try that some time.
It is also very good and cheap for computer use.
Matched roughly with Sol on DeepSwe cost per task.
Luna way cheaper. DeepSeek used to be, but I think it's somewhere on Sol's curve after the price hike.
On DeepSwe it's strictly beaten by Luna on max, cost and result.
Damn, Luna on max is as good on DeepSWE as Kimi k3, I think I dismissed this model unjustly.
This is why benchmarks are scary, since Artificial Analysis puts it a fair bit behind Kimi K3.
Kimi K3 is a beast though, just costly.
1 reply →
So in a way Luna is the new Gemini Flash? I've been out of the game for a while.
1 reply →
Why is Gemini represented by points on this cost-quality plane, while competitor's models are represented by curves?
Competitors release multpile models and their curves reflect reasoning effort of each single model.
Gemini doesn't have adjustable reasoning effort (at least on the graph) so each of its curves is just one point.
Gemini has levels though
That graph has to be made because an Exec didn't like that graph went down to the right instead of up and to the right. How do you make a graph with 0 on the far right and counts up by going left of 0? What number line is that?
For what it's worth the source of the data[0] does have 3.7 flash with all 3 reasoning levels. 3.5/3.6 are in fact just the single points though (high reasoning). The datapoint in the announcement screenshot is either med or high, but they're pretty much exactly the same so can't say for certain.
[0]: https://deepswe.datacurve.ai/
It is extremely odd that this model has that crowbar, spending more for worse results at "high".