Hi, I’m on the AI Pro plan and I’m seeing some very confusing behavior with the Antigravity model quotas for Gemini Pro.
Yesterday I barely used any agents or did meaningful work with Antigravity. This morning, when I started Antigravity again, my Gemini Pro quota was suddenly at 80% instead of 100%, and the dashboard now says it will only reset about 42 hours from now. To double‑check, I looked at my recent activity using the “Antigravity Cockpit” extension, and its logs also show almost no usage for yesterday.
Because of that, it feels like the system applied a weekly quota deduction overnight, even though my actual usage was very low.
What I would have expected
Quota to decrease in proportion to my real usage.
No large weekly penalty if I’m hardly using the service.
A short cooldown (the advertised 5‑hour refresh), not a ~42‑hour wait, for a Pro subscription.
What I’m seeing instead
When I opened Antigravity this morning (2026‑03‑12, JST), the Gemini Pro remaining quota was already down by 20% (100% → 80%) compared with the previous day.
The reset time is shown as about 42 hours from now.
My recent activity, confirmed via Antigravity Cockpit, does not justify such a large deduction or a multi‑day cooldown.
Additional details
Plan: AI Pro
Model: Gemini Pro
Time zone: JST
The Antigravity quota UI history shows the 20% drop appearing between yesterday and this morning, not gradually with usage.
Could someone from the team clarify whether this behavior is expected (for example, due to some hidden weekly cap or batch accounting), or if there might be a bug in how quotas are being calculated for my account? If this is expected, it would be very helpful to have the exact rules documented in the product so users can understand when and why such multi‑day cooldowns occur.
I just went through the same thing. I woke up to find my baseline quota had been reached. I’m not the type of person who pushes things that hard, and I definitely checked yesterday to see if my quota was still there.
I’ve been looking into this, and while it feels like a bug, there is a high probability this is related to how Agents and the Weekly Baseline interact in the background.
If you are using agents for complex tasks (like code refactoring or multi-step reasoning), here are a few “hidden” reasons your quota might have vanished overnight:
The “Marathon” vs. “Sprint” Quota: The Pro plan often tracks a Weekly Baseline (your long-term pool) alongside the rolling 5-hour limit. If you ran high-intensity agent tasks yesterday, the “Marathon” baseline might have been depleted. Because the dashboard syncs are sometimes delayed, the “0% remaining” status often doesn’t update until the system reconciles your total token usage overnight.
Agent Recursion & Context Bloat: A single prompt to an agent isn’t just one request. If an agent is assigned a task that requires it to “loop” or self-correct, it can trigger 10–20 internal model calls. If that agent is also passing a large codebase or a complex DOM as context with every step, your token consumption scales exponentially.
Infinite Loops: It’s possible an agent got stuck in a reasoning loop or a “retrying” state on a background task. If it was still processing when you closed your laptop, it could have chewed through your baseline while you slept.
Background Triggers: Check if any of your agents are tied to automated triggers (like GitHub Actions or API webhooks). These can fire and consume quota without you ever opening the chat interface.
Recommendation: Check your Agent Manager logs to see if there was a spike in “Internal Calls” or “Tokens Consumed” during your off-hours. Switching your background agents to Gemini 3 Flash for simpler tasks can help preserve your Pro quota.
Thanks for the detailed explanation. I’d like to clarify my situation a bit more, because based on how I’m using Antigravity, I still don’t see how I could have hit any weekly baseline limit.
I’m almost using Gemini 3 Flash, not Pro, for my agents.
When I’m not actively using Antigravity, I close the IDE, and I make sure all sessions are finished before closing it.
I’m not running long “refactor the whole repo”–type jobs, background workflows, or multi‑agent setups.
I also don’t have any agents wired to external triggers like GitHub Actions or webhooks.
For this week in particular, my Antigravity usage has been very light. The quota dashboard was at 100% the whole time, so I don’t see how small increments across the week could have silently accumulated into a weekly‑baseline hit. It feels unlikely that a few short Flash‑based sessions would deplete enough of the weekly pool to cause a 20% drop and a 42‑hour lockout the next morning.
To double‑check, I reviewed my recent activity using the Antigravity Cockpit extension, and I don’t see any spikes or long‑running jobs there either.
One more question:
Are quotas for the Antigravity IDE shared with other Gemini entry points such as:
the Gemini web UI,
Gemini API usage (Code Assist, CLI, etc.), or
tools like Jules?
If these all draw from the same underlying quota bucket, that might explain the behavior. But if they are supposed to be separate, then my case still looks more like a bug or a miscalculation in how the weekly baseline is applied.
One more thing I noticed, which might be relevant:
I also checked a different AI Pro account (my work account), and it shows a similar refresh pattern. In that account, all model quotas are currently at 100%, but the next refresh times are:
Gemini 3.1 Pro (High/Low): refreshes in 7 days
Gemini 3 Flash: refreshes in 5 hours
Claude Sonnet / Claude Opus: refresh in 7 days
This makes me wonder if the allocation or refresh rules for the AI Pro plan have recently changed, for example to a 7‑day “weekly” window for non‑Flash models.
I’ve attached a screenshot of the quota page from that second AI Pro account as well for reference.
I have the exact same problem. I did use it yesterday for a tiny task that would have never exhausted the quota. Everything was fine before I went to sleep, now it suddenly says that the model quota is reached and I have to wait 7 days. Honestly, Gemini with the pro plan has been quite exhausting for me with their frequent quota changes. I now have reactivated my Codex plan. OpenAI seems to be much more generous with their limits and their new models are pretty good.
same issue here! its not the first time the session just spikes after night (pc was turned off, nothing running). but this time my gemini lost 40% (and reset window from 5h to 5 days!) and my claude went from 60% to 0%!
i am paying for PRO and i did age verification as i saw suggested on another post..
makes me seriously consider moving to claude platform.
added screenshot of the gemini pro usage.. those spikes are not making any sense.
AI Pro Plan here, pretty much the same situation.
I spent gemini 3.1 pro to ~20%, then it shows need 2 days to refresh instead of 5hrs.
What even worse is, when I passed that 2 days without using it, instead of refresh to 100% it dropped to 0% and another 7 days for “refresh“.
Very strange: yesterday it was at 100%, today I launched Antigravity and it immediately showed 80%. Then I restarted it and it was already at 60%. I hadn’t even done anything. What the hell is going on?!
I’m also planning to switch to Codex soon. I’ve heard from many people that the latest models are extremely good for coding and automation, so I’m excited to start experimenting with it and see how it can improve my workflow.
Yeah same here as everyone. Tuesday and Wednesday left the Claude Model at 60% and Gemini at 20% according to AGCockpit extension. On Tuesday i would like to add that i saw Claude drop from 60 to 40 after….. a few hours of not using it and the reset time of quota reach, but i didnt give it much tought.
Wednesday 0 usage… i didnt even open the IDE.
And today i found 0% on both.
I sent an email to antigravity-support@google.com
And i think they restored the full quota because i saw my quota in a 100% for all models (Claude and Gemini were on 0% after 1 day of 0)
Same for me, I have to use antigravity for at least 4-5 days with my personal pro account, opened it just right now to do some work and saw all quotas exhausted. Doesn’t even worth paying 3 euros per month for this anymore
