Which premium model to use

How To
1

Helloouu everyone,

which models are You using for Composer at this moment?
which one works for You and why?

in chat or editing instructions (ctrl+k) - different model ? lighter one?
cursor small is really well and fast with small adjustments or small questions in my opinion

What are Your thoughts?

DeepSeek compared already to other models or maybe too soon?

There are also Gemini models i see - anyone using?

Claude3.5 sonnet and Claude3.5 sonnet 20241022 - which one is the latest one?

Also the Claude Haiku or Opus - what could be the use cases for them?
usage is different with premium requests

i came from VS Code so this is a new world for me

EDIT:
forgot to ask - cursorsmall - is it based on other LLM?
is around 3B model?

if topic already exists - my appologies!

Thanks for reply
and have a good coding time!
R.

2

I would use Claude 3.5 sonnet for everything. It works great and it feels like the software was designed around it.

Other models can be helpful for specific use-cases, but 99% of the time I’m using Claude.

Today I’m using Gemini-2.0-Flash, and I find it to be an improvement, but it doesn’t support Composer Agent so I’ll probably switch back to Claude until its supported.

3 Likes
3

Yep! thought so!
bute then Claude3.5 sonnet and Claude3.5 sonnet 20241022 - what is the difference? can’t find too much probably the 20241022 is the latest?

4

From

They are identical currently I believe as 20241022 is the latest.

  • Claude3.5 sonnet points to the latest every time.
  • 20241022 is referencing a specific version released in November.
5

Opus is supposed to be the better high level thinker. But I don’t think it works in Agent mode currently.

Seems Opus might be a better agent, and Sonnet a better file writer.

I think something we need in Cursor is ability to change which model is used for which tasks.

I personally would love both the ability to set a default model for task X, but also a list of acceptable ones with tradeoffs defined, so that the agent can make a different choice if needed.

6

Sonnet 3.5 for best Agent performence.
Gemini 2.0 Flash or Deepseek V3 for smartest performence in free section.
Haiku 3.5 if you want to maintain more Fast request. Haiku only uses 1/3 fast request so u get 1500 instead of 500.