Claude Sonnet & Opus 4.6 Tokens Being Consumed While Using Gemini 3.5 Flash (High)

Google Antigravity
, ,
1

I am currently working on a project in Google Antigravity and using Gemini 3.5 Flash (High) for most of my interactions.

However, I noticed that my Claude Sonnet and Opus 4.6 token usage appears to be decreasing even when I am not actively using those models. My conversations and queries have been directed to Gemini, so this behavior is confusing.

Could someone clarify:

  1. Why are Claude Sonnet and Opus 4.6 tokens being consumed when Gemini is selected?
  2. Is there any background processing or routing that uses these models?
  3. How can users accurately track model-specific token usage?
  4. Has anyone else experienced the same issue?

If token balances decrease without explicitly using a model, it becomes difficult to trust the usage reporting and platform transparency.

Any clarification from the Antigravity team would be appreciated.

2

Hi @Vishwesh_Gaur,

Welcome to the forum!

  1. Ideally, only the selected model’s quota should be consumed.
  2. Antigravity does not perform any background processing that would use a different model.
  3. Since all Gemini models share the same quota pool, and all non-Gemini models (such as Opus and Sonnet) share a separate quota pool, you can currently only check the quotas consumed by these broader model types, rather than by a specific model.
  4. We haven’t been able to reproduce this unexpected behavior based on your description. If you are still facing this issue, to help us better assist you and investigate further, could you please provide a few additional details?
    • Which Antigravity platform and version (e.g., 2.0, IDE, CLI) you are using
    • A sample prompt you used
    • Whether this behavior is observed across all projects or only specific ones
3

I have experienced this too, it seems very odd. Wasn’t sure if its a bug or intentional tokendemonics’ :frowning: