I used exclusively Composer 2.5 fast. model and it drained 100% of the “API limits” and now draining the “Auto + Composer limits” - it looks like a bug, I thought the “API limits” are only used by the other models like GPT-5.5, not by Composer models.
I had Run Mode set to “Auto-review”
I had the explore sub-agent model set to Composer 2.5 fast
In the Cursor spending table, all I see is Composer 2.5 fast
So there’s nothing that could have really use the API limits.
Composer 2.5 should draw from your Auto + Composer pool before it touches your API limits. Once that pool is exhausted, any additional requests will fall back to the API pool.
In the Usage tab of your dashboard, you can adjust the timeframe being reviewed. The default is set to 7 days, which may not cover your entire billing period. Adjust the timeframe to see the full picture, and don’t forget there might be multiple pages of results.
Hey @Colin it means there’s definitely a bug in Cursor. As I mentioned, it drained the “API limits” first. I was literally watching the “API limits” being drained daily, while using exclusively Composer model. I know you can see spendings for the whole month, it shows 99% of “Composer 2.5 fast” requests and a few very small “GPT-5.5” requests. I thought it was kind of weird, but I didn’t really care since it’s almost as good as GPT-5.5. But still, it’s a bug
I might be mistaken and 10 of my GPT-5.5 requests (~1 mil tokens) could actually drain $60 subscription limits but that’s unlikely. And this also doesn’t’ explain me literally seeing how the composer requests draining the API limits in real time (about 1% per 1-3 requests), I remember checking it multiple times because it didn’t look right
I’ve tried to reproduce this and I haven’t been able to. On an individual Pro account, Composer 2.5 uses up the Auto + Composer bucket before moving to API usage.
I can see that your usage limits just reset. How are things looking in the usage dashboard now when you use Composer 2.5?
Hey @Colin thanks for checking. I might have been mistaken about this being a bug, the last suspect I have is the commit review automation that was draining the “API limits” automatically on commit push in the background, but they are also displayed in the spending history and I see there’s not that many of those requests. Perhaps the automations just drained the $60 sub quicker than I thought they would.
Perhaps it would be a good idea to somehow display that info about automations and their spendings and runs more clear, maybe even in cursor IDE somewhere in chat area so users don’t drain their limits so quickly and don’t forget there’s some things setup in the background?
Similar problem on macOS
Version: 3.9.16 (Universal)
When I explicitly select Composer 2.5 with Fast disabled, usage events show composer-2.5 + Included, but:
Auto + Composer pool stays at 0%
API pool / Total spend increases with every request
This happens with onlycomposer-2.5 events visible in usage
Expected (per docs + support):
Composer 2.5 should consume the Auto + Composer pool first. API pool should only be used for third-party models (Claude, GPT, etc.) or after Auto + Composer is exhausted.
Sample events (Jul 5, 2026, ~19:37–19:42 CET):
Time
Status
Model
Tokens
07:42 PM
Included
composer-2.5
90.6k
07:41 PM
Included
composer-2.5
351.1k
07:40 PM
Included
composer-2.5
489.6k
07:37 PM
Included
composer-2.5
792.1k
After these requests: Auto + Composer = 0%, API bar increased.