I have been trying to run deepseek v3 0324 and gemini 2.5 pro on bolt.diy and 98% of the time, i am not getting any results. It’s either it freezes, errors or some other thing. Can bolt. Diy be optimized or made to be as fluent as bolt.new?
Hi @Victorkind1,
thats provider/model related. You tried 2 very need models and the system prompts maybe not working really good with them.
If you use e.g. Anthropic with Sonnet 3.7 (same as bolt.new), it runs pretty smooth/fluent.
Having the possibillity to choose whatever model you want, has also disadvantages, which you can see here. As there is no standard how models output their stuff exactly, it would need system prompts etc. per Model to get best results. Planned for the future is a prompt library, so the community could provide better working systems prompts per Model.