There is a model usage tracking/billing misattribution on the web dashboard. When a session started with Composer 2.5 spawns a Composer 2.5 Fast sub-agent, the resulting usage is logged and billed as Grok 4.6 (high) instead of the actual model used.
Steps to Reproduce
Open Cursor and start a new Composer session with the model set to Composer 2.5.
Run a task that triggers the main agent to automatically spawn a sub-agent on Composer 2.5 Fast.
Allow the sub-agent to complete the task/requests.
Go to the web dashboard and check the Usage Details page for the corresponding session timestamp.
Expected Behavior
The usage incurred by the sub-agent should be recorded as Composer 2.5 Fast (or tracked under the appropriate Composer tier).
Hey @Parksi, thanks for the detailed report with screenshots. I can see both Usage Details and the agent window.
I checked this session. Your Composer 2.5 session and all sub-agents it started were recorded and billed as Composer 2.5. None of this session was billed at the Grok 4.6 rate.
The Grok 4.6 (high) lines in Usage Details are from a separate Grok Bot session that was running on your account at the same time. Grok Bot uses Grok 4.6 by default and shares usage with your Cursor plan, so its activity shows up next to the Composer session by timestamp. This can make it look like one switched into the other. Similar cases were discussed here: Is Grok Bot usage separate from Cursor plan? and Grok Bot spend cursor usage, I can't accept it
One small note about the tile. The sub-agent showed up as Composer 2.5 Fast, but the run actually used regular Composer 2.5, which is cheaper, not more expensive.
Can you check whether you had Grok Bot open or had any Grok Bot automations set up around 12:15 to 12:45 UTC on August 31? If you did not start Grok Bot at all, reply here and we’ll dig deeper.
Hey @deanrie , thanks for checking this so carefully.
You’re right — I did have Grok Bot open around that same window (roughly 12:15–12:45 UTC on Aug 31), so the Grok 4.6 lines almost certainly came from that session, not from the Composer 2.5 agent run. Appreciate you confirming that the Composer session and its sub-agents were billed as Composer 2.5.
That said, I’d really like to request a UX / usage-display improvement here. Right now it’s extremely easy to confuse the two:
Grok Bot usage and IDE / Composer Agent usage show up in the same Usage Details list, interleaved only by timestamp
There’s no clear product label (e.g. “Grok Bot” vs “Cursor Agent”) on each row
In my case the agent UI also showed “Composer 2.5 Fast” while the run was actually regular Composer 2.5, which made the mismatch look even more like a billing bug
Would it be possible to separate Grok Bot usage into its own section/filter, or at least tag each event with the product source? That would have avoided this whole false alarm for me, and probably for others too (given the related threads you linked).
Happy to close this out as “not a billing mischarge” if the team can track the clearer labeling as a product improvement.
Glad we cleared it up. Confirming there was no mischarge: the Composer session and its sub-agents were billed as Composer 2.5, and the Grok 4.6 high lines came from a separate parallel Grok Bot session.
On your suggestions, this is solid feedback, thanks for laying it out clearly:
Splitting or filtering Grok Bot vs Cursor Agent usage in Usage Details, and showing a clear product label on every line. I’ll pass this to the team as a product improvement. I agree it’s easy to mix up right now, and related threads show it’s not just you.
The tile showing Composer 2.5 Fast when it actually runs regular Composer 2.5. This is a known display issue and we’re already tracking it. I can’t share an ETA, but it’s on our radar.
I’m closing the billing part as not a mischarge. If there’s an update on labeling, I’ll follow up. Let me know if anything else comes up.
