Gemini 3.6 Flash(console.cloud.google.com) |
Gemini 3.6 Flash(console.cloud.google.com) |
Judging from my own frustrating attempts to use Gemini for vibe-coding, it seems like Google is badly over-sold (i.e., under-provisioned).
From all those promos giving away their pro-level subscription with phones; spinning up a mid-level subscription to undercut other providers and (probably most significantly) putting AI queries into ever search response because their flagship search product had become useless; they're promising a lot more processing to customers than they can reliably deliver.
The recent iterations seem to be intended not to push the capabilities forward, but to deliver capabilities at the current level while consuming less resources. That will allow them to maintain their trajectory until (I'm expecting) they get the huge infusion of extra compute resources from Space X later this year.
If I'm right, then I expect we should see Google start pushing forward again (rather than more of this lateral stuff) by the end of the year.
Based on my experience lately with Codex, it seems the opposite. Gemini 3.6 Flash (high) in Antigravity CLI feels a LOT faster than GPT 5.6 Luna (high) in codex. In the order of 10-50x.
There's no way these are using a significant amount of compute. I'm not 100% convinced they're actually LLM-generated rather than an old-school Markov model. Both the relevance and accuracy numbers of the responses flirt with 0%. It's possible they just have a few million stashed responses and choose one at random, from what I can tell as a user.
https://blog.google/innovation-and-ai/models-and-research/ge...
What is interesting is how this is announced before any Gemini Pro progress. From the outside it seems as though Google cannot keep up with other frontier models.
Flash 3.6 is certainly capable of many coding tasks, but clearly a model of this size of not trying to compete at the frontier as a software development tool, and not clear why Google really need to complete there other than for PR-related AI bragging rights.
I don't know how a frontier model like GPT 5.6 or Fable could have done better (I have no need that justifies paying for them), but yesterday I used the free Gemini chat app (i.e. Flash 3.6) to discuss and explain this poorly written recent AI paper to me, and honestly couldn't ask for much more.
Edit: Oh it's in the other link
https://blog.google/innovation-and-ai/models-and-research/ge...
They're very well equipped to be the volume discount store of inference.
(Quoting myself from 2 days ago.)
Gemini 3.6 Flash (High)
Gemini 3.6 Flash (Medium)
Gemini 3.6 Flash (Low)
Gemini 3.5 Flash (Medium)
Gemini 3.5 Flash (High)
Gemini 3.5 Flash (Low)
Gemini 3.1 Pro (Low)
Gemini 3.1 Pro (High)
Claude Sonnet 4.6 (Thinking)
Claude Opus 4.6 (Thinking)
GPT-OSS 120B (Medium)Aren’t Marketing/PR people at FAANG supposed to be among the best in the industry?
[1] https://www.lexiconbranding.com/case-studies/google-antigrav...
so I think they're just focussed on releasing whatever helps their bottomline (improving search?). its not like anthropic and openai are making a lot of money being on the top of charts! this might just be my crazy pills talking though.
releasing a model worse than luna is pretty bad. Its clear that internally they did not decide coding was a thing until relatively recently.
Gemini 3.1 Pro just feels a generation behind, from when models would miss easy things and make bad assumptions. Its not actively detrimental in bad way but the opportunity cost vs using something like Opus to be productive is large.
Gemini 3.5 Flash is the most annoying model I have ever used. It loves to respond in ALL CAPS like “LOOK AT THAT” for no apparent reason. I realize it’s a flash model but I will give it a basic list of tasks and the output will simply vomit “Now I will X” “Now I will Y” “Now I will Z” over and over again filling my screen with garbage. It’s also no smarter than 3.1 Pro and consumes just as many tokens as 3.1 Pro, it’s really pointless without a newer Pro model in place.
3.1 Pro: while it's coding performance is mediocre, a lot of coding work requires minimum actual thinking. I use it often for light refactoring, boilerplate generation, testcase skeleton generation, code review, language questions ("is there a better way to write this code block?"). I don't have a corporation behind me so costs matter. Considering that I'm a Pro subscriber, it's quite cost effective.
3.5 Flash: excellent model for general (re)search. I use it for everyday tasks with Thinking instead of Flash-Lite. It's a much better version of Google Search for general queries like gaming tips, cooking, day-to-day first aid, tax and investment questions, etc.
Google is clearly aiming for cost-benefit here and considering that it gets bundled with YouTube Plus and Google One at $20/month, it's a killer deal IMHO.
PS: I don't work for Google and don't even like Google very much. But this is a good product.
Sonnet is roughly the same level as 3.1 pro for me.
rumor has it that pro is not so pro in internal evaluations and that there are internal stakeholders holding it up
A couple of days ago I was comparing Claude vs Gemini asking both what the size of YACC generated parsing tables for ANSI C was. Claude was very gung-ho and rushed off to download a C grammar, install bison, and find out for itself (not really necessary), and then hit daily free usage limit half way through without even getting to an answer (despite this being the only thing I'd asked it that day).
Google's "AI Overview" search results, which started out awful, are now much improved - can be actually useful - as long as you are not asking specialist questions, and the Gemini chat and voice apps are also great for everyday use, although not sure if they are yet monetizing this (volume probably a lot less than search I'd guess).
The Apple Siri-Gemini deal is also a significant way for Google to not get sidelined.
Around here we'd come to that conclusion at least a couple of years ago, due to abusive SEO and so forth. And that understanding was becoming even more widespread.
I don't know if Google's got the later parts of the game figured out yet, but I have to think that they'd realized that Search was dying. While there's still some value there, may as well use it as a hook to pull people into what they expect to be the next era.
Last time Google had a "everybody agrees" SOTA model was Gemini 3 in November 2025, it beat GPT-5.1, it really was better and held it for a month.
Currently Google's best model doesn't match open models, in any category (performance, price or speed), in fact the last 4 open weight champions all beat Google's best model.
Here's a plot of the evolution: https://www.reddit.com/r/LocalLLaMA/comments/1v20g29/kimik3_...
I have no opinion on whether this is true, but "It's been 6 months since Google had the best model in the world" seems a rather weak criticism!
It's anyways been clear for a long time that people are finding value at all sorts of different model sizes and price points, and that Pareto frontier and cost-to-complete task are more important than who benchamaxxed who.
If a model as strong as Gemini 3.6 Flash(!) had been released a year ago, then everyone would be falling over themselves calling it AGI - it is extremely capable, and free usage in the chat app is essentially unlimited.
At least OpenAI and Anthropic can still tell their customers their best model is better than some Finnish student can setup in the customers' basement, even if the difference is pretty small now. But they are better. Google is not.
Surely that's a major change in the situation. "America" is losing the advantage (although even the Chinese models are losing, see next paragraph) but Google is a big step behind 6 other players and 3 open models, so let's just say, Google is effectively behind everyone else that matters. Worse: one of the models that beats Google's best model was trained on Chinese ASICs (that aren't even 2nm. And very few I might add, couple thousand. Google's next model ... has a LOT to prove)
Even Chinese models are losing their advantage. By which I mean that if you check how far Qwen 3.6 27B is behind trillion-parameter models, it's less than a year, down from easily 3 years. If that continues to go down, even Chinese trillion parameter models will become hard to justify (and Qwen 3.8 is being cooked up as we speak, I mean we don't even know if a new 27B is in the works or not, but everyone's very excited)