I upgraded to Google AI Ultra and started building a new app in Antigravity. After about 5 hours of development using a thinking model for complex math and geometry, I got rate limited and then effectively locked out from using the model I paid for. The IDE downgraded me to Flash, and in my case it looks like I’m blocked from returning to the higher tier for up to 7 days.
I get that agentic IDE workflows are more expensive than normal chat. One “request” can trigger planning steps, internal checks, tool calls, retries, and long reasoning traces, and heavy math problems can spike compute. I also understand rolling windows and protective caps exist.
But here’s the issue: at $200 per month, this experience doesn’t match expectations. I’m not buying a trial or even a pro experience. I’m buying reliability for real work. If a normal evening of development can trigger a multi day lockout, Ultra doesn’t feel workable for its intended audience.
What I think needs to change
-
Real quota transparency inside Antigravity
Show an actual quota meter for the IDE that reflects agentic compute, not just a plan name.
Show remaining budget for the rolling window and when it meaningfully recovers.
Show estimated cost before running heavy planning tasks. -
Higher and more predictable limits for Ultra
If Ultra is positioned as premium, the ceiling for sustained dev sessions should be higher and more consistent. If compute cost is the constraint, publish the limits clearly so people can plan. If task is chewing compute for 3-5 minutes maybe try an alternate approach -
Better degradation behavior
A silent downgrade to Flash mid workflow can break development because capability and behavior change. At minimum, make the downgrade explicit with options, and avoid multi day lockouts that strand people. Maybe even finish the task you are in the middle of as a courtesy. I had a long running task that I was confident was going to fix an issue I had but hit my limit before it finish its long thought process.
Practical note for others
If you’re doing complex math, geometry, or heavy refactors in an agentic IDE, the compute burn seems dramatically higher than expected. That’s exactly why transparency matters.
Question for the team and other users
Is it expected that Ultra users can hit a 7 day style lockout after about 5 hours of development in Antigravity? If yes, what are the actual quotas and reset rules? If no, what’s the right escalation path to get account level throttling reviewed?