Deepseek v3 beats sonnet 3.5

We’re constantly evaluating models and we’ll add them when they surpass our existing model range on our internal benchmarks!

While DeepSeek might be working very well for your use case, we only want to add the model to Cursor if it is equally well-performant in everyone’s use case.

release a beta version for enthusiasts please, parallel branch with new stuff

Any updates on this?

At least make it easy for us to switch between deepSeek and other models. Having to change the cursor settings and toggle the API key every time is really frustrating.

Any news? I want

  1. Custom API with my own key that doesnt disable builtin sonnet. Maybe with bigger limits like 20k context??

  2. Deepseek3 that only uses 1/5 of fast request (since its 30 times cheaper that Sonnet). 1/5 sounds reasonable even after DS3 price increase in a month

2 Likes

If you say he is not even as good as the 4o-mini, the conclusion is obviously absurd

2 Likes