I love Cursor, but the performance of Auto Mode is making me consider switching back to Copilot. Once I reach my plan’s limit, the fallback model is almost useless. It writes code like an inexperienced junior developer, forcing me to spend hours fixing mistakes. The Cursor strategy team should reconsider their ‘limit’ experience; otherwise, churn will definitely increase.
I also confirm that the Auto model has very disappointing performance.
Cursor Auto Mode seems to be using the Composer model for most things, and that model is TERRIBLE in following rules and command instructions. Not even to mention the bad code generation.
I 100% agree. It does not follow instructions at all. Earlier I selected sonnet 4.6 and it was so much better. I am absolutely surprised by its poor performance
The only model I thought was perfect was Composer 1.0. From 1.5 upwards, it’s become garbage; it’s impossible to work with.