Usage quotas as of last month or so has made Gemini AI Pro with CLI and Antigraivity worthless

The AI Pro tier is functionally broken for development right now. Since the release of version 0.38.x, a paid subscription has turned into a series of rolling lockouts and depleted quotas that make it impossible to finish a workday.

The biggest issue is a massive regression in the CLI agent logic. When signed in via OAuth (the standard Google One login) the CLI now has a hard cap of 200 requests on a 24 hour rolling window. This is a significant downgrade from the 5 hour refresh available before the 0.38.x update. Prior to switching to an API key to test the limits, just 3 or 4 prompts burned 36% of the entire daily quota in 15 minutes. In a recent session on a small repository, two targeted prompts resulted in 76 total requests. The agent is spending most of its time in background utility loops, talking to itself and re reading files instead of executing changes.

Antigravity is equally unreliable. One or two prompts can immediately spike usage to 20% of the available quota. There is clearly a hidden scaling system or weekly baseline in the backend that ignores the promised 5 hour refresh. If you hit a high token interaction just once, you are penalized as if you have spent a massive portion of your total capacity.

Testing this with an API key shows the actual cost of this inefficiency. A 15 minute session for a simple function fix cost over $2.00 in raw credits because of the millions of tokens burned by background agent chatter. This happened despite using a targeted prompt for a finite task, proving the agent is racking up massive input costs through redundant internal turns.

Google is marketing a high request limit for AI Pro, but the current architecture is so inefficient that it burns through that capacity in minutes. We are paying for a “Pro” tool that is essentially DDOSing our own accounts with redundant background tasks. Until the agent stops spawning dozens of internal requests for small tasks and the 5 hour refresh is actually honored, this subscription is worthless for a professional or even a hobbyist workflow.

Google are making the Pro tier useless
the bottom line tier is Ultra
even Ultra users are complaining about token burning in side tasks rather than actual work, and other says even failed prompts are consuming tokens.

Just experiencing the same thing this weekend May 9th and 10th, if I get locked out after 2 or 3 requests (on Anti-Gravity) then have to resort to “Gemini 3 Flash” then… whats the point in the Pro subscription?

oh and my baseline quota to be refreshed in 5 days !?!? locked out until then ?

this change or quota thing just mucked up my whole day.
since I could not use gemini 3.1 Pro (low), I used GPT-OSS 120B, to conclude the work done today…

well that lead to total FUBAR in the branch… a simple change resulted a cascade of required changes, then I reverted back, and functionality was broken , so I quite GPT OSS and went to Gemini Flash… and just running around in circles trying to restore the functionality I had working before changing to GPT OSS

I now spent more time running in circles than actually implemention working functionality…

Gemini 3 flash seems as clever as a rock…

Thanks!