Bolt Forge

Bolt Forge is an agent inside Bolt.new, the AI app builder, that runs on open-source AI models only. It launches on September 14, 2026 as a research preview and appears as a third agent in the agent picker alongside the Standard and Max agents that Bolt users already work with. Every individual Pro plan receives up to 50X more usage at no extra cost through October 14, 2026, which lets builders draft, test, tear down and rebuild without rationing prompts. Forge is designed for people who arrive at Bolt with an idea and the skill to see it through, and who want room to experiment widely before spending premium credits on polished production work. Price has decided who gets to build with AI at speed and scale. Builders arrive at Bolt with an idea and the skill to see it through, then ration prompts like fuel: every brainstorm, every rough draft, every dead end burns credits priced for polished work. The drafting phase of building, the part where you need room to wander, is exactly the part usage caps punish. Forge removes that penalty. The allocation is big enough to draft, test, tear down and rebuild, and it is where builders can test a big idea, play around with new concepts and experiment as far outside the box as they want to go. A second motivation sits on the model side. AI inference costs have dropped 280x in 18 months (Stanford HAI AI Index, 2025), and that collapse is what makes an allocation this size possible. Open-source models also improve when they see how real software gets built, the one thing that cannot be scraped. Opted-in Forge sessions provide it, with consent. Builders get room, and open models get better. Forge runs open models that builders can see and select. As of September 2026 that means GLM 5.3 Flash and GLM 5.3 alongside it, with Kimi K3 and DeepSeek v4 Pro as experimental options. The lineup will evolve, and when a new open model drops, Forge is the first place in Bolt it lands. Bolt documents two honest caveats. These models are experimental inside Bolt, so the guidance is to duplicate a project before switching a serious build into Forge and to keep complex production work in Standard or Max. The models also do not all cost the same to run: Kimi K3 and DeepSeek v4 Pro burn through usage faster than the GLM pair, so starting on the default and reaching for the heavier options when the work calls for it is the recommended path. Forge also cannot take PDF uploads yet. Capability was tested before shipping. The Forge lineup ran through the Bolt Build Index, Bolt.new's own benchmark for how well a model completes real Bolt projects, and came out at 91% of the top paid model's score (Claude Opus 5) as of September 2026. In raw numbers, Forge's open models score 92.2 against 101.0 for the top paid model in Bolt.new. The trade-off is nine points; the payoff is up to 50X more usage at no extra cost on every individual Pro plan. Bolt runs Forge on its own reserved hardware instead of paying a provider per request. A fixed, predictable cost means more of what a builder pays goes to building instead of markup. Forge projects run in the browser on WebContainers, the technology StackBlitz built and Bolt runs on, so no server is rented for every build. Builds stay fast and costs stay low. Between the reserved hardware and the browser runtime, the cost structure is the reason Forge's price works. Every individual Pro plan includes the Forge allocation at no extra cost through October 14: up to 50X more usage, drawn from a single monthly bar that resets on the renewal date. Every Pro tier gets the same allowance, Forge usage is separate from Standard and Max usage, and there are no daily limits. When the bar hits 100%, Bolt switches the builder back to Standard rather than charging an overage, so there is no daily math and no surprise pause mid-project. Training in Forge is opt-in: a one-tap consent screen spells out the trade in plain language before anyone builds in Forge. The shared data is prompts, code, project files and configuration, the tool calls Bolt makes, and edit histories including the fix traces Bolt creates. Bolt de-identifies shared sessions before anything leaves its infrastructure, stripping secrets, sensitive data and personal information as a rule, and validating the pipeline against seeded test data. The sessions go into datasets that Bolt licenses to AI developers under a data license agreement. Arcee AI, a U.S. open-model lab behind the Apache 2.0-licensed Trinity model family, is the first. Bolt has partnered with Arcee to help train a trillion-parameter-class model, and sessions builders share during the preview window, September 14 to October 14, 2026, feed the first training run, which begins in October. A data license agreement governs every transfer, StackBlitz may be paid for the datasets it licenses, and those datasets may go to other AI developers as well as Arcee. What makes the corpus valuable is what is in it: the models learn from the work people do in Forge and only there, starting projects, making edits, connecting databases and fixing errors. A model that has seen the job gets it right in fewer tries. The workflow is five steps from start to finish. First, pick Forge in the agent picker, where it appears as a third agent next to Standard and Max on every individual Pro plan. Second, opt in with one tap after a consent screen that spells out the trade in plain language. Third, build, drawing Forge usage from one monthly bar with no daily limits and a hard stop at 100%. Fourth, Bolt de-identifies shared sessions before anything leaves its infrastructure. Fifth, the sessions go into datasets that Bolt licenses to AI developers, starting with Arcee AI. Switching back to Standard or Max stops Forge from collecting anything new, and builders who want Bolt to stop using Forge content it has already collected can email privacy@stackblitz.com. The benefits are framed as what the trade buys. Brainstorms, MVPs and experiments stop competing with production work for premium credits. Builders get 91% of the top paid model's Bolt Build Index score, included with Pro at no extra cost. Usage is readable at a glance: one monthly bar, no daily limits, and a hard stop instead of an overage bill. And builders get a hand in what comes next, because their sessions teach open models how real software is actually built. In practice that means scenarios such as drafting and tearing down an MVP, exploring new concepts outside the box, running throwaway experiments while premium agents stay reserved for production work, and contributing consented build sessions that feed the next generation of open models. Forge is aimed at individual Pro builders in Bolt.new. Teams and Enterprise workspaces are excluded from Forge, and from AI training and dataset licensing. Sessions from the EEA, the UK and Switzerland are not used for training or datasets, so builders in those regions get the Forge allocation without the trade. Pro is $25 a month billed yearly and includes up to 50X Forge usage at no extra cost with no access code required. Builders not on Pro can join the Bolt Lite waitlist, with codes going out in waves and the first seats opening by September 21, 2026; Bolt Lite costs $9 a month and anyone on it when sign-ups close on October 14, 2026 keeps the plan at that price. Forge is a third agent beside Standard and Max, so premium agents, allocations and privacy settings stay exactly as they are, and builders can switch between agents at any time. After October 14, 2026 the research preview window closes, but Forge keeps going as an open-model lab: the up-to-50X allocation on Pro and the first training-data window with Arcee AI both run from September 14 to October 14, and what comes next depends on what the preview shows. Forge is experimental inside Bolt, so serious production work belongs in Standard or Max. The takeaway is the trade at the centre of the product: builders trade de-identified, opted-in build sessions for up to 50X more usage on open-source models that score 91% of Bolt's top paid model. Builders get room, and open models get better.