Here’s an opinion that I have, that Google is actually providing much better limits than most people claim, and if you are consistently hitting the quota, you need to adjust your workflow.
Here’s the current pricing on Gemini Flash 3 (Cheaper than Provided Gemini Flash 3.5)
With a $20 Pro plan, including 5 TB storage, Veo, enhanced Gemini and other tiny features, we should get at least $15 of usage to make it worth it. $20 should be preferable.
Based on my own experimentation, we can get ~50m input / 20m output per month (I only recorded for a single month).
In other words, Google is providing each user with 60-80$ of quota for only $20 USD. The quota is actually much higher than cursor and almost double that of copilot (making sense as copilot is half the price).
I think I had much more in January. In March or April, I realized that it goes down several times. Now it’s worse. The data is definitely there.
I don’t know, maybe Google doesn’t want to show this data?
I think Antigravity and Google AI Studio have different computations. Just thinking… Antigravity used to be like it’s unlimited during Feb-May, but now with the frequent updates nerfing usage, it seems useless even for pro users. In my experience, the past month I never reached the quota even if I’m using Antigravity for a whole day straight, but now a few prompts and a single complex task can exhaust the models (the 5 models at the same time, even though the low model is selected, idk why).
What happened is that Google gave us several times the expected limit for marketing purposes. But obviously, they can’t just keep pouring the money into a product and decide to cut down everything to reduce their costs
Some people bought annual subscriptions because of Gemini-cli, where you could easily do 500 million input every day. Antigravity is anti-performance on any context, after all, you can’t even load 600k tokens at the start there, in the config.
Gemini-cli allowed loading as much as you wanted into the context at the start via settings.json. That was its strong suit, we could work on specific data without any issues. Antigravity completely removed this, how should I put it, basic and damn necessary feature. That’s exactly why I got the annual subscription.
I doubt that Cursor or Copilot have this. They are optimized for coding, not for working on full context in one-shots.
The model probably has it, but that doesn’t change anything. Because in Gemini-cli you could have data loaded automatically into the context, without wasting requests. Currently, just loading such a file in batches of 400 or 2000 lines once got me a 3-day ban.
AI Pro is unbeatable as value. 5TB Google Drive, Youtube Premium Lite, FAMILY plan, video creation. It covers everything your entire family needs for AI plus all the extras.
Even Antigravity limit is not bad at all, easily as good as a dedicated $20 account for Claude Code and Codex.
Sure Antigravity is not quite as polished as Claude Code and Codex, Veo 3 is not as good as Seedance 2. But the overall value is simply unbeatable and Google quality is already good enough, and will only get better.
Yeah, the only problem right now is that some people don’t understand how to think around it. Some users underestimate the power of the models. Gamini 3.5 Low is awesome for creating implementation plans and writing code. For audits I prefer to use Medium to High reasoning Claude.
Exactly! Google is improving fast, and will increase the limits once it becomes possible. Right now, they have slashed enough to turn a profit and they would try to increase it from now on
Hey there! I hear you, and I wanted to share some thoughts on the recent changes.
Firstly, we used to have ample Gemini-3-Flash quota before May. The current concerns really started with Gemini Flash 3.5 and the removal of the Flash quota pool. Plus, Claude model delays of 2-4 days seem to be a quiet weekly limit cooldown, which is a bit of a surprise.
Also, direct Gemini API usage (e.g., in Cursor) taps into different quota pools than Antigravity does right now. But soon, on June 18th, the new Antigravity CLI/IDE will merge all these pools. This means users transitioning might find their effective quota reduced or costs increasing.
Thirdly, Gemini-3.5-Flash is becoming more autonomous, sometimes taking unprompted actions. This can increase the cost per response. While quotas might seem higher, this increased consumption can quickly balance it out. If you prefer to steer the workflow, these extra steps can add up!
In summary, it seems less about ‘limits are excellent’ (especially without unlimited traffic! ) and more about the lack of clarity. Unlike Cursor, where token values are clear, Pro/Ultra plans don’t specify exact token volumes or costs. This unpredictably shifts the value we get for our money. It’s like visiting your favorite restaurant and feeling unsatisfied, wondering if the portions shrunk or the bill is off.
My point remains that the current limits are much higher than cursor or Claude, and compared to Copilot. It also remains that you are receiving the value for your money.
For lack of transparency, I agree but we aren’t even discussing it
Firstly, we used to have ample Gemini-3-Flash quota before May. 
The current concerns really started with Gemini Flash 3.5 and the removal of the Flash quota pool. 
Plus, Claude model delays of 2-4 days seem to be a quiet weekly limit cooldown, which is a bit of a surprise.
Also, direct Gemini API usage (e.g., in Cursor) taps into different quota pools than Antigravity does right now. 
But soon, on June 18th, the new Antigravity CLI/IDE will merge all these pools. This means users transitioning might find their effective quota reduced or costs increasing.
Thirdly, Gemini-3.5-Flash is becoming more autonomous, sometimes taking unprompted actions. This can increase the cost per response. 
While quotas might seem higher, this increased consumption can quickly balance it out. If you prefer to steer the workflow, these extra steps can add up!
) and more about the lack of clarity. 
Unlike Cursor, where token values are clear, Pro/Ultra plans don’t specify exact token volumes or costs. This unpredictably shifts the value we get for our money. It’s like visiting your favorite restaurant and feeling unsatisfied, wondering if the portions shrunk or the bill is off. 