Why does cursor keep setting model back to auto when using my own api key?

Support Help
,
1

I set the api key, saved it, and updated the model to gpt-5.6-luna and started “coding”. Randomly, the model switches back to auto and now I’ve spent $10 in on demand for no reason.

Is there a way to stop this from happening? It’s $10, not the end of the world, kind of slimy behavior.

“He’s not paying attention, switch back to auto so we can get some money!” I know that’s ridiculous but that’s what it feels like.

image
image1104×791 102 KB

6

Hey, thanks for the report and the screenshot, it clearly shows what’s happening. The User API Key lines with $0.00 confirm your key was working correctly while gpt-5.6-luna was selected. Charges only show up on the On-Demand / auto lines.

What’s actually happening, there’s no bad intent here. When you hit the “Other Models” limit, Cursor automatically switches the model selection to Auto. Auto is designed to never use your personal API key, so requests go to Cursor models, and since your included limit was already used up, they get billed as on-demand. The fact that this switch happens silently while BYOK is enabled shouldn’t happen, and it’s not anything you configured. I’ve flagged it to the team.

Until this is fixed, there are two ways to stay safe:

  • Set your on-demand spend limit to $0 in the Dashboard. Then when you hit the limit, requests will pause instead of charging you.
  • If you see a notice like “Switched to Cursor Models / Other Models limit reached”, just re-select gpt-5.6-luna.

About the ~$10 charge, for an account-specific review, email [email protected]. A separate team handles this and they can see billing details.

Let me know if the switch happens again even with the limit set to $0.