I created a prompt for my CRM development and tested it with both bolt.new and bolt.diy. The bolt.new platform provided a fast preview, while it took approximately 30 minutes to get a preview on bolt.diy.
With what LLM on which provider did you test?
claude sonnet 3.5 New
Ok, can you share your prompt and/or export chat, so we can verify on our instances?
Thanks for chiming in and helping @leex279!
@parthsolankipd, try using Claude 3.5 Sonnet Old instead of New. For some reason the new model doesn’t work the best with the Bolt prompting in my experience. It hallucinates incorrect code handling the React import and usage while the old Claude never does.
I believe the commercial Bolt.new is either still using the old Claude 3.5 Sonnet or they adjusted their prompt to prevent these weird issues with the new Sonnet.
Thank you for your quick response, I used Claude 3.5 Sonnet Old and it work awesome. You should create a topic and pin post “Consider these points before using bolt.diy”
3 Likes
Yeah I appreciate the suggestion! What I’m going to do right now for sure is update the troubleshooting pinned post to talk about models to use!
1 Like