Rate Limited but drawn from wrong pool?

Google Antigravity
, , ,
1

So before I get into the background, someone can maybe explain this to me… I selected Claude Sonnet 4.6 in AntiGravity, and I’m now rate limited while at the final step or two (gradle failed because of debug keystore, claude was fixing that when i got locked).

antigravity-error
antigravity-error1571×1277 147 KB

Background: I’ve spent the last 2 days working on my app in AI Studio. What started as a joke has actually become something good. one of the central points that make it stand out though is LiteRT-LM and Gemma 4 on-device. Started with Gemini 3.5 Flash Preview. Wanting it to implement LiteRT-LM model loader and download either the gemma-4-e2b-it.litert or gemma-4-e4b-it.litert files. simple, right? especially since its a google platform. It created an entirely fake, fraudulently represented implementation. check out the fake log messages.

(i couldn’t put this image here because i’m limited to one embed per post, but the photo is posted in the AI Studio forum in my post there… gemini wrote code that falsely claimed that it was implemented, wrote fake inference interactions, and then wrote fake log messages claiming they were generated locally offline with gemma 4 e4b, temperature 0.7, and either 110ms or 320ms depending on what it felt like choosing. it was very very intentionally deceiving me)

So I switched to Gemini 3.1 Pro Preview. It did some stuff but generally didn’t implement it. so today I switched to Gemini 3.1 Pro Latest, and it told me twice that it was fully functional and no issues with the code. After it still not working and 2 days of time wasted, I exported the code via zip and loaded it up in AntiGravity. my first time using AntiGravity. I had only ever used Gemini and Gemma AI (i’m sheltered, i know) so I decided to give the Claude Sonnet 4.6 model a try. In 10 minutes, it fixed what 3 Gemini models couldn’t do in 2 days time with Google’s own AI Edge platform. But you guys wouldn’t let him finish, because for some reason, even though I only hit than 67% on claude, i’m maxed out on Gemini, so Gemini cut me off from Claude?

Make this make sense. I can’t see continuing with Gemini Pro…

2

Hi @technotic,

Based on the screenshot you provided, it looks like you have reached your 5-hour limit, rather than being cut off by Gemini. The quotas for Gemini models and Claude/GPT models are actually tracked in completely separate buckets.

What you experienced is the intended behavior of the dual-limit system:

  • You still have 67% of your weekly limit for Claude remaining.
  • However, you have completely exhausted your 5-hour limit (hitting 0%) for the Claude/GPT group.

This 5-hour burst limit acts as a temporary pause to fairly distribute server capacity. Whenever it hits 0%, the weekly limit doesn’t apply until the short-term timer resets. Once your 4-hour and 44-minute timer finishes, your 5-hour limit will refresh, and you can dive right back into your remaining 67% weekly quota to finish your app.