bolt.new
Prompt, run, edit, and deploy full-stack web apps
Start new thread

Bolt Forge - Bolt's new agent with open-source models and 50X usage

by•
Bolt Forge is a new agent inside Bolt.new that runs on open-source models. Every Pro plan gets up to 50X more usage with no daily caps. Scores 91% of Bolt's top paid model on their internal benchmark.

Add a comment

Replies

Best
Hunter
📌

Bolt Forge is a new agent inside that runs entirely on open-source models.

Sitting next to Standard and Max in the agent picker, this one is built only on open models - GLM 5.3 Flash and GLM 5.3 as the main pair, Kimi K3 and DeepSeek v4 Pro as experimental picks.

What makes it different: Every individual Pro plan gets up to 50X more usage on Forge, free, through Oct 14, 2026.

The trade is opt-in: your build sessions (prompts, code, fix traces) get anonymized and go toward training new open-weight models with Arcee AI, a US open-model lab. You can stop sharing anytime by switching back to Standard or Max.

Key features:

  • One monthly usage bar, no daily limits

  • Hits 100% → auto-switches to Standard, no overage

  • Scores 92.2 vs 101 for Bolt's top paid model on their internal benchmark (~91% capability)

  • Runs on WebContainers in-browser, no server cost

  • Duplicate your project before testing serious builds, still a research preview

Who it's for: Builders who burn through credits fast in the brainstorm/draft phase and want room to experiment without touching production usage.

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified →

You, and people like you, have killed this platform. I really hope you're proud of yourself. This sort of automated hunting is a big part of why Product Hunt feels so different now. Racing to post other people’s launches, paraphrasing the announcement and then adding a follower CTA turns product discovery into personal audience farming. Makers should be the centre of the post, not the hunter.

Incredible work from the Bolt team. As far as I know, it's the first of the vibe-coding solutions to really go all in with open-source models, and I commend you!

"91% of Bolt's top paid model on their internal benchmark" is doing a lot of work in that sentence - internal benchmarks tend to be picked to flatter the new thing. Is there a public eval or a task set outsiders can rerun, or is 91% something we just have to take on faith for now?

Question for the makers—right now it handles full-stack in the browser tab really well, but what’s the vision for connecting to existing local microservices or private databases during the prototyping phase?