these gpt 4o-mini / cursor-small requests were being incremented when our apply model was used (for legacy usage accounting reasons). the main chat/composer response was still from whichever model you selected in the app
just pushed a fix to stop counting apply usage here, let me know if you still see this!