Possible Bug in Cursor Token Statistics: Other Models Disabled After Grok Quota Exhausted

Hey, thanks for the detailed report. This isn’t a bug. It’s expected behavior, but I agree it’s not obvious.

Your two pools aren’t fully independent:

  • Named third-party models (DeepSeek, GPT, etc.) only pull from the Other Models pool.
  • Cursor models (Auto, Composer, Cursor Grok) use the Cursor Models pool first, and when it’s used up they automatically continue from the remaining Other Models pool.

That’s what happened here. Heavy Grok 4.6 usage first drained your Cursor Models quota, then it spilled over and finished off Other Models too. That’s why both show exhausted. Before, Composer 2.5 looked like it was billed separately just because you never fully used up the Cursor Models pool, so the spillover never kicked in.

This two-pool behavior is described here: Models & Pricing | Cursor Docs

What you can do now: your quotas will reset at the start of the next billing cycle (you can see the exact date in the dashboard), or you can enable on-demand usage in settings to keep going until the reset.

Related threads if you want to dig deeper:

Let me know if anything’s still unclear.