So I use gpt 4.1 for everything I do. Somneone told me that Auto is just GPT 4.1? Is this true and are there any differences at all?
Auto dynamically switches between many models including Claude Sonnet 4.0.
Auto is absolute trash. How these developers at anysphere even thought it was a good idea is beyond me. Each model has different layers on how it interprets context. In one session, I can see I’ve been switched numerous times, entirely different generation styles, each model loses context of a piece of text literally sent about 10 seconds prior. Sometimes you get a model that is “Okay”, other times you get absolute junk responses. It’s just outright terrible.
5 Likes
There’s literally no chance it switches between sonnet 4. The results speak for themselves.
The experience is insane…the way context is lost is crazy. At the very least, we should know what llm we are using at a given time. Maybe we work out a system of llms that are ok to work with and can ask to rotate between them. They also make it so that you cant find out which llm you are working with. i will be researching self-hosting an llm as consistency is more important to me than speed.
Ask it what model it is using. It does sometimes use sonnet 4. Open 20 instances of cursor and ask each one what model it uses.
1 Like
Auto-generation is great for quick conversation and fast answers. For coding, though, it’s better to use Claude or other top-tier models.