StackBlitz's browser-based AI development platform Bolt.new is offering developers up to 50 times more usage of open weight coding models in exchange for anonymized training data from their work sessions. The company launched Forge on Monday as a research preview for individual Pro subscribers, running through October 14. Developers who opt in will share anonymized versions of their sessions—including prompts, source code, fix traces, and conversations with the coding agent—to help train a trillion-parameter-class open-weight model in partnership with Arcee AI.
The program gives developers access to GLM 5.3 Flash and GLM 5.3 models, with Kimi K3 and DeepSeek v4 Pro available as experimental options. Forge's open models scored 92.2 on Bolt's internal Build Index, compared with 101.0 for the company's top paid model, reaching about 91% of the top score. The first training run is scheduled to begin in October, with Bolt saying the resulting model weights will eventually be released publicly. The 50-fold usage boost ends October 14, though Forge will continue as an open-model testing ground after the preview period wraps up.
Forge requires an explicit opt-in, with a consent screen appearing each time a developer switches into the workspace. Standard and Max sessions aren't included, and Teams and Enterprise accounts can't participate. Bolt says it anonymizes sessions before they leave its infrastructure, removing secrets, sensitive data, and personal information, and tests the process against seeded data. Arcee receives the resulting data under a signed processing agreement. Developers can stop sharing new sessions by leaving Forge, but anything already used for training will remain in the models.
The program reflects a broader industry bet that coding trajectories—the record of what worked and what didn't during development—matter more than finished code alone. Public repositories contain enormous amounts of source code, but they mostly show the end result, while a coding session can fill in the gaps left by failed attempts and revisions along the way. According to the report, that record becomes useful as coding agents take on longer jobs that require finding the right files, coordinating changes, and recovering when something breaks—especially when agents inherit code written by other agents. Forge changes the underlying setup by running open weight models on Bolt's own infrastructure, with the company's WebContainers technology providing a cost advantage by running projects in an isolated environment inside the user's browser rather than on Bolt's servers. If Bolt can run more of its coding workloads on its own models instead of paying for proprietary APIs, it gains more control over costs and usage, while the Forge sessions help train whatever comes next. The arrangement mirrors a similar approach from SpaceXAI last month, which trained Grok 4.6 on agent failure traces—the missteps, retries, and corrections that other labs typically discard—though Bolt is sourcing the data from developer sessions rather than synthetic runs. This trade-off between free compute and proprietary training rights may redefine how developers value their workflow data, while platform economics increasingly favor companies that can close the loop between user activity and model improvement.