What LLM is best for coding

bolt.diy [bolt.diy] General Discussion
1

So far I had been using bolt.new which is using Claude under the hood if I recall well. This worked very well and code quality is great.

With Ottodev I have been using various other LLMs, but to be honest, nothing as good as Claude.

So I’m wondering what are the options I have now:

  • When I purchase an Anthropic key (18$) can I have the same experience as Bolt.new?
  • Any other models worth trying out for coding?
  • I heard good stuff on QWEN for coding, but the issue is that I only have a nVidia 3060 card. I could host it on Runpod, but that’s quite expensive

Any other options?

1 Like
2

Great questions!

Claude 3.5 Sonnet is still the best for coding right now, only to be trumped by o1 once that is available through the API for the general public. Though it certainly isn’t as fast!

Qwen 2.5 Coder 32b and DeepSeek Coder v2 236b are my favorite open source LLMs for coding. You definitely need beefy hardware to run them yourself, but you can use a provider like OpenRouter to use them through an API for dirt cheap! WAY cheaper than using Claude.

3

Hello, it’s really great to be here. I would like to know if you could direct me to a list of coding llm’s and a tutorial of how to add them to Bolt.diy? This would be really helpful. Thank you in advance.

1 Like
4

for online ones you just select from the list and add your api key(s). for local ones you can use ollama or lmstudio. do a google search for each if you need instructions, but it’s dead simple. there are also lots of yt vids.

5

for local ms’ phi-4 is (supposedly) at the top now.

1 Like
6

phi-4 is good - but its only 16k context windows(how much the model can read and write in one pass). Context is crucial to keep the model understanding all the dependencies and the code. I am starting to have some good success with the new Gemini Flash with 1M context myself. I used to be on that team - so I am a little bias:)

2 Likes
7

oh wow that’s awesome abt gemini and also that you were on that team!

i’m quite sure a ver of the model will show up on hf with an increased context size

1 Like