Where does the bug appear (feature/product)?
Cloud Agent (GitHub, Slack, Web, Linear)
Describe the Bug
I wanted to try Grok 4.5 on cloud agent, and when I did, I noticed that on the usage page it was showing Claude 4.5 Sonnet.
I tried stopping it and sent a follow-up message, and it still used and billed for Sonnet 4.5.
Steps to Reproduce
Starting a cloud agent from cursor.com/agents with Grok 4.5 as the model
Screenshots / Screen Recordings
Operating System
Other
Version Information
It’s Cursor Cloud Agent
Does this stop you from using Cursor
No - Cursor works, but with this issue
@deanrie Can you please check? I want to try out Grok 4.5 on Cloud Agent, but I’m not sure if I’ll be billed for Sonnet 4.5 again.
Hey @risonsimon
That single “Claude 4.5 Sonnet” label almost certainly doesn’t mean your whole run was billed as Sonnet.
A cloud agent run usually uses more than one model under the hood. The model you pick does the main work, but it also spins up helper sub-agents and occasionally falls back to another model. Right now the usage page collapses that whole run into one representative model label, so a Grok 4.5 run can show up as “Claude 4.5 Sonnet.” The tokens and cost shown are the combined total for the run, not proof it all ran on Sonnet. This is a known display issue we’re tracking. Both of your rows are under “Included,” so this came out of your included usage rather than extra on-demand charges.
To confirm exactly what your agent used and make sure nothing was mis-billed, could you drop the agent link (the cursor.com/agents/... URL for that run)? I’ll trace it on our end and sort out a credit if anything’s off.
This is the agent link: https://cursor.com/agents/bc-835277f4-8e4f-4540-a83f-7910cc642b9c
That’s interesting. I’d assume subagents would use Composer, not an older and more expensive model like Sonnet 4.5.
Also, is there a way to control the model used for subagents? I want to predict the costs.