Bolt.new launched Bolt Forge on September 17, 2026, a research preview that gives individual Pro subscribers expanded access to open-source AI models in exchange for opt-in sharing of anonymized coding sessions. The data goes to Arcee AI, the Western open-weight lab behind the Trinity model family, which will use it to train a trillion-parameter-class open-weight model.
Forge is a selectable mode in the model picker and runs frontier open-source models only. At launch it supports GLM 4.6 Flash, DeepSeek, and Kimi, and Bolt.new says new open-model releases will be added as they drop. Every individual Pro plan now includes 50X more usage on open models at no extra cost. Usage inside Forge appears as a single monthly bar with no daily limits and a hard stop at 100 percent.
The research preview period runs from September 14 to October 14, 2026. Bolt Forge is live today for individual Pro subscribers, and more information is available at bolt.new/forge.
Consent Is a Tap, Not a Clause
Consent is obtained through a one-tap opt-in each time a builder switches into Forge. Bolt.new anonymizes shared sessions before data leaves its infrastructure, stripping sensitive and personal information, and the anonymization pipeline is validated against seeded test data. A signed data processing agreement governs every data transfer. Builders can leave the program at any time. Teams and Enterprise workspaces are excluded from Forge and all data collection, and premium modes remain private and untouched.
The company frames the arrangement as a trade with a visible receipt. Every session from the first Forge training window contributes to the model, and the receipt for the trade is the model itself. The circle closes where the weights publish: what builders help train, they can download, inspect, and run, like anyone else.
Simons: Arming Builders Is the Point
Eric Simons, co-founder and CEO of Bolt.new, described the launch as a bet on open-source models as the way to widen who gets to build software. Price has been deciding who gets to build with AI, he said, and Forge opens AI building to far more people.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
"With Forge, you forge your own products and businesses," Simons said. "And by building, you're helping the world forge capable open-weight models. A year ago, a lot of these builders didn't have the purchasing power to do this at all. Arming them is the whole point."
Arcee: Capability Stops Being Something You Rent
Mark McQuade, co-founder and CEO of Arcee AI, said the partnership rests on ownership rather than access alone. Models trained on Forge data will be released publicly for download and local running.
"When a model's weights are open, capability stops being something you rent," McQuade said. "Models trained on Forge are released to the public, so anyone can download and run them, including the builders whose work made them better. Widening who gets to build software starts with widening who owns the models, and open-source models belong to everyone."
A Two-Sided Exchange With a Public Output
The launch addresses cost as a barrier to AI software building, pairing wider model access for individual subscribers with a public release of whatever the sessions help train. Builders get expanded usage, and the public gets open-weight models. The arrangement is presented as mutually beneficial rather than one-directional: the builders whose sessions feed the training window are also the people who can download the finished weights.
For Bolt.new, the preview tests whether opt-in data sharing can sustain a model-training pipeline at a scale individual subscribers can reach. For Arcee AI, it adds a stream of anonymized coding sessions to the work behind its Trinity model family. Both companies say the privacy measures are built into the product rather than buried in terms, and that the exclusions for Teams and Enterprise workspaces keep business data out of the program entirely.
The preview runs for a month. What comes out of it, by both companies' description, is a model anyone can run.