Possibility to set temperature

It would be a great feature to have the ability to set the ‘temperature’ for the request on chat window. I use 0.3 as it is the sweet spot, where the model does not “hallucinate” but gives always proper code.

curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
     "model": "gpt-3.5-turbo",
     "messages": [{"role": "user", "content": "Say this is a test!"}],
     "temperature": 0.3

Yeah, I agree, it’ll be great if we could change the model parameters like temperature or top p and …
it happened to me that I got the proper reply when I messed around with those params
@moderators , what do you think?, thanks in advance for having a look here


Fully agree, just wanted to make a post about it but it’s already been requested in September.
If the API that is compatible with the OpenAI v1 chat completions framework, it should be the made available imo. Should not take more than a few hours to implement I think. @truell20 This would be a great addition.


Any updates on this?

A few of our features depend on temperature=0 for working well (in particular for speculative edits).

We’ve found that temperature 0 is best for higher-quality suggestions, so we don’t plan to support this in the short-term.


I’ve been using Cursor daily for nearly a year and overall have been quite happy! Recently I’ve started looking for alternatives because of this though.

I’d prefer to be able to entirely opt out of Cursor prompts as well. Over 50% of the time I leave cursor to use other solutions because quality is much higher. I suspect if you set temp to zero your prompts are what are causing this.

Other than just temperature, having all parameters exposed is really necessary for me. No cursor written prompts, settings, etc. If it breaks some features that’s okay with me. Having the ability to opt out is necessary, at least for me.

Sad that it sounds like this isn’t on the roadmap.

1 Like

I’ve noticed this too. The performance of Opus in Cursor isn’t impressive, and GPT-4o’s also leaves a lot to be desired. It often struggles. So, I decided to run my problem on Antropic’s Console or the GPT-4o Playground with 0.3 (Which is the sweet spot for Coding i read), where I can adjust the temperature, and both managed to solve the problem I was facing in a single attempt.

This needs implementation because it’s affecting code generation, leading to more Fast Requests being used, which is quite literally a waste of money at this point.


Just curious, do you have a meta prompt or some kind of context when you try Antropic Console or GPT-4o like this?