Hey, thanks for the report. This is expected behavior, not a bug.
On the Free plan, only Auto mode is available, and it uses Cursor’s Composer 2.5. Named premium models like GPT 5.6 Sol aren’t available on the Free plan, even if they show up in the picker. The request still goes through Auto or Composer. This is covered in the docs: Available models | Cursor Docs Paid plans unlock all models.
To use GPT 5.6 Sol, you have two options:
Upgrade to a paid plan, or
Add your own OpenAI API key BYOK in Cursor Settings > Models.
Thanks for the extra detail, that changes things. My previous reply was about the Free plan because your forum email shows as free, but on Pro you do have access to the GPT 5.6 Sol model, so the Included charge isn’t related here. The selected model should run as chosen, the request status doesn’t affect that.
Most likely this is caused by Auto mode. If Auto is enabled in the model picker, Cursor will pick the model itself, often Composer 2.5, no matter which named model is highlighted. Please check this, turn off Auto and explicitly select GPT 5.6 Sol before sending, then see what actually gets used.
I do not have AUto enabled. I specifically use ChatGPT and see that Cursor is starting to use Composer without my permission. Last month I saw similar problems, Cursor started to use MAX mode, though I have not enabled it.
I have a similar issue. I am on a Pro plan. I select Composer 2.5 or any other model (auto is off, multitask is on) and cursor completes the task always using ChatGPT 5.6 no matter what model I selected and it counts against my ‘Other Models’.
Hey, thanks for coming back with details. If Auto is off and the selected model still isn’t being used, that shouldn’t happen, and it’s not your setup. The previous reply about the Free plan doesn’t apply here, you’re on Pro.
To dig in, we’ll need a bit of info from each of you:
The Request ID from a session where the wrong model ran right chat menu in the top-right corner > Copy Request ID. If Privacy Mode is on, the Request ID might be unavailable, so please turn it off temporarily and retry the request.
Are you using multitask or subagents? @Seb_Wallat, you said multitask is on, that’s important. With multitask, work can be handed off to a subagent, and that subagent can sometimes run on a different model than the one selected in the main picker. @KES, are you using multitask or subagents too?
Your Cursor version and a screenshot of the model picker right before you send the message, so we can see what’s selected and that Fast is off.
With the Request ID, we can check the path the request took, confirm which model actually executed, and pass it to the team to review the routing. This is an issue we’re tracking, but the exact details depend on specific Request IDs. I’ll reply here once there’s an update.
