Which is the Best Model to Run?

Title: Which is the Best Model to Run?

I tried the pro plan on bolt.new, but this month specifically, I spent a lot of tokens compared to the same workload last month. I think they changed the model they’re using because the current model on bolt.new is very, very dumb. Last month’s model was much smarter.

Currently, I’m trying bolt.diy with deepseek, but the results have become worse. It’s cheaper in price, but the results aren’t the best.

What I want exactly is to work with last month’s model on bolt.diy because it was better, and with diy, it would be cheaper in cost.

Welcome @ElJoker,

bolt.new uses “Claude Sonnet 3.5” and maybe they switched to the new one. I testet these days as well with bolt.new and got the same impression as you. Doest work very well. Did not pay for it but spent the 200k free token with at the end no valuable output.

If you like to use the old model from bolt.new you can e.g. use OpenRouter with “Antrhopic Claude Sonnet 3.5 (old)”

In the FAQ are some Models mentioned: Frequently Asked Questions (FAQ) - bolt.diy Docs

I personally like Google Gemini 2.0 and the new experimental and they are free to use.

I built this application mostly with it:https://www.youtube.com/watch?v=FMejkH8aOFE&list=PLz0wwBxxxGRyb9CpzjCFXPh10COWdg4Zp

I would recommend using 0.0.3 Version of bolt.diy at the moment, because we are seeing some bugs in the latest versions and bolt not doing well, what is frustrating at the end for the user. We hope to fix/find the issue soon.

2 Likes

Can you share the OpenRouter link with “Anthropic Claude Sonnet 3.5 (old)”?