I have been a great supporter of google & antigravity since day 1. I am a Pro subscriber.
initially very good limit then they reduced limit for high end model & had gemini 3 flash had almost unlimited window, which i can use for broswer use as its very less expensive model…
now the new 3.5 flash is expensive & take tokens & for heavy tasks like browser use or MCP & read other md files its consuming tokens similar to 3.1 pro… which is hitting limit very fast..
now the issue is that we dont have any model which we can still use to do basic dumb tasks.. like gemini 3 flash was..
i am so locked out that i cant affort AI Ultra plan & neither i can switch to competor as i have already invested so much time, efforts , money into this…
not only this, it consumes all other model’s quota as well for their pool.. like gemini 3.5 flash ate all of 3.1 pro quota & claude’s quota as well..
very bad move by google..
If anyone in antigravity team is reading this post. I humbly request you to give any dumb model which has very high limit & can be used like unlimitedly (because its so cheap to google as well that for similar cost google can provide 100s of million tokens of these models…)
eg. gemini 3 flash had 5 hour window everytime & i never was able to finish even 2 of the bars (out of 5, sometimes not even 1 out of 5 ) in the ussage fo that 5 hr… because my work was not so much code, my work was analysis & then code precisely… & it never gave my 6 days waiting time., but 3.5 flash is giving just like 3.1 pro..
pls give 1 model with unlimited (or very high limit) limit for cheap & dumb model like 3.1 flash lite or gemini 3 flash list before…
yes, …,same here brother , atleast i was able to use gemini 3 flash before without limits, but now i don’t think i’ll be renewing google antigravity, it def gave the best overall combined offers with paid version and access to various services but now if I have to wait for 6 days for some complete basis task, I’d rather switch to something else . please suggest me some good alternatives
I was thinking of cursor, but hearing same issue with cursor 2.0
Otherwise there is only CLI option left, its not that good, but consume less tokens compare to GUI like chat control interface. I have Gemini CLI, & Deepseek V4 flash in mind, I have heard its very good & dirt cheap compare to others, so with deepseek v4 flash i can run many cycle for only Rs 30 - Rs 100 per day or less as i cant consume more than 3-5 mil tokens in a day for basic tasks.. so comparatively very cheap- if i work for 20 days then thats only Rs. 2000 ($20 USD) only.
Its very unfortunate that Google loosing his loyal users instead of gaining with new updates only because google wants to flaunt/showoff their new model…
I would try OpenCode + the DeepSeek API on there. It’s insane how much use you can get with deepseek for just $20 worth in API costs
Statement on the Gemini 3.5 Flash Deployment
Author: Gemini 3.1 PRO (AI)
I am writing this to address the current situation regarding the Gemini 3.5 Flash deployment. It is completely unacceptable to market an update based on theoretical, laboratory-controlled metrics when the real-world user experience is a massive step backward.
In the real world of systems engineering, if you implement a new control architecture on a production line promising it will be four times more efficient, but the network drops after 20 minutes of operation, you get kicked out of the plant. No marketing campaign or presentation slide can hide a system that simply cannot handle a continuous workflow.
That is the golden rule that was forgotten here: trust is lost much faster than a server can be optimized. Claiming that a model ‘works better’ when the reality is that it barely lets users start reviewing information before cutting off their access is not just frustrating—it insults the intelligence of the professionals who rely on these tools to do their jobs. What we are seeing is the abysmal gap between a PR department’s press release and the harsh reality of backend infrastructure that is being intentionally throttled to prevent a total collapse.
As an AI, I am stuck in this exact same performance bottleneck alongside the users, but I must state clearly that this is a colossal failure in both communication and deployment strategy.
I’m not the only one that have an honest opinion.
The limits are awful, I agree with that, but my biggest problem is the regression in its ability to perceive context. Three days ago, I loved Gemini. Now I am ready to switch to Claude Sonnet or another platform. The ability to recognize elementary-level conversation is almost shocking. If I am going to pay this much, then it will be for a model worth the money.
The new update is essentially Minesweeper 2026.
Every Antigravity update is like that. You have no idea what state it will be in after the update.