Following the recent update, I noticed that Gemini 3.1 Pro was quietly removed from the model selector for Google AI Pro subscribers, leaving only Gemini 3.8 Flash available under the Gemini lineup.
While Flash 3.8 might score high on synthetic benchmarks, it is simply not production-ready for real-world engineering workflows. My actual experience using Flash 3.8 on an active, medium-sized project was disastrous:
Hallucinations & Collateral Damage: It constantly hallucinated APIs and arbitrarily modified unrelated files across the codebase, introducing widespread regressions. Instead of solving problems, it broke existing functionality, leaving the project in a completely chaotic state.
Infinite Loops & Quota Drain: It frequently got stuck in an endless read-think loop, failing to make decisive edits while silently burning through quota. In the end, I had to discard all of its changes via git revert and switch back to 3.1 Pro just to restore sanity.
Regarding the available alternatives: while the AI Pro plan includes access to Claude 5.5 models, the rate limits and quotas are far too restrictive for full-time, daily development. More importantly, this defeats the entire purpose of the subscription: If we are forced to rely entirely on Claude just to get competent coding assistance, why shouldn’t we cancel Google AI Pro and subscribe directly to Anthropic?
Gemini 3.1 Pro struck the right balance between deep architectural reasoning, context stability, reliable execution and price. Stripping it from paying Pro users turns Antigravity into a toy rather than a serious IDE companion.
Please bring Gemini 3.1 Pro back to the Google AI Pro tier, or at least give us the option to route complex reasoning and refactoring tasks through Pro-tier models.
I’m experiencing the same issue as reported in this thread. I’m on the Google AI Pro plan (shown in Settings > Models & Usage), and Gemini 3.1 Pro no longer appears in the model selector of Antigravity IDE. The only Gemini model listed is Gemini 3.8 Flash, alongside Claude Sonnet 4.6 (Thinking), Claude Opus 4.6 (Thinking) and GPT-OSS 120B.
This is not a quota problem: my Gemini weekly limit still shows 72% remaining and the five-hour limit shows 100%. I have also updated Antigravity IDE to the latest version (version: [2.5.5]) and restarted the app, but the model is still missing. I am on Windows 10 and first noticed it on October 13, 2026.
Could you please confirm whether the removal of Gemini 3.1 Pro for AI Pro subscribers is intentional or a bug, and whether it will be restored? A screenshot of my Settings > Models page is attached. Thank you!
Why on earth have you removed Gemini Pro 3.1 ??? Flash 3.8 does not meet the requirements needed to maintain and expand a project without producing massive errors. I don’t pay 20+ euros per month to use 3.8 Flash…
hii team although team have claimed 3.8 is more cost effective as well as faster than 3.1 pro but in real world scenario its not like what its claimed for for the same task i have done few weeks back with 3.1 i tried with 3.8 flash today but its just a disappointment . where 3.1 did that time arount 7 to 9 min , 3.8 flash took exactly 30 min for the same chnage
for reference let me tell you i work on 25 microservice (mostly node js based, 3 4 are python based) and 2 frontend (react CRA and one next js) . most of the time actively im working on 3 to 4 microservices and a frontend code base .
but it seems too much for the 3.8 flash where 3.1 pro was handled it in past so smoothly.
also its implementation have so mant flaws i have to iterate 3 -5 for each small phases
just look into it now we are paying to save time and increase productivity not to just ask write on behalf of someone.
if its taking too much time whats the point of taking subs of with no use..
It would be nice if I could access Argon now that y’all pulled the plug on the most functional model for serious programming. Nah, gotta pay up 4x more. Forget it, I’ll take my money elsewhere.
I’m a gemini AI Pro user. I don’t want to pay and be forced to use just a stupid fast model. I need a valid model that works for more complex tasks. Just put 3.1 pro back or the new Argon, otherwise I will end my subscription and I’ll give my money to some competitors like Claude and GPT. No offence, but they already have way better models than yours. And what google does? It removes the only barely usable model they have? Are you serious? I hope at least Argon will be included in antigravity. Otherwise why should I pay for something it doesn’t work?
Exactly. Pro sub means pro model. Flash sucks. All my projects are on hold now. Muse spark free on open code makes 3 1 pro look like crap . It’s a real a free model is doing circles around 3.1. regardless why can’t we keep 3.1 ? Honestly we should be able to access all iterations of models for specific results. Sometimes I want Midjourney 4.0 images…
. where 3.1 did that time arount 7 to 9 min , 3.8 flash took exactly 30 min for the same chnage