Bolt Forge Gives Coders 50x More Free AI Power For Your Build Data

Bolt.new launched a mode called Forge on September 14, 2026 that gives every individual Pro subscriber up to 50 times more usage on four open-weight coding models — GLM 5.3, GLM 5.3 Flash, Kimi K3, and DeepSeek v4 Pro — free through October 14, in exchange for opting in to have AI-assisted build sessions used to train a planned trillion-parameter open-weight model with Arcee AI. On Bolt's own Build Index, Forge's open models scored 92.2 against 101.0 for Bolt's top paid closed model, roughly 91% of frontier performance. Bolt also opened waitlist sign-ups for Bolt Lite, a $9-a-month plan built on Forge's open models, with access rolling out from September 14 and sign-ups closing October 14, the same day the free usage boost ends.

Bolt.new is giving away up to 50 times more usage on open-weight coding models through October 14, free, and the price is your build data. Bolt.new turned on a new mode called Forge on September 14, 2026, and the trade behind it isn't hidden. Opt in, and your AI-assisted builds help train a planned trillion-parameter open-weight model. In exchange, every individual Pro subscriber gets up to 50 times more usage on four open-weight models, free, through October 14. The models on offer are GLM 5.3, GLM 5.3 Flash, Kimi K3, and DeepSeek v4 Pro. Kimi K3 and DeepSeek v4 Pro burn through usage faster than the GLM pair. So Bolt defaults new users to GLM and lets them reach for the heavier models only when a job actually needs them. Here's the number that matters. On Bolt's own Build Index, Forge's open models scored 92.2. Bolt's top paid, closed model scored 101.0. That's roughly 91% of frontier performance, running on models anyone can download and host themselves, not a proprietary system only Bolt controls. Say yes to Forge, and your session gets scrubbed of anything personally identifying, then routed to Arcee AI, the open-weight lab behind the Trinity model family, including the 400 billion parameter Trinity Large. Bolt and Arcee are aiming their first training run at October 2026, working toward a model in the trillion-parameter class. A signed data processing agreement governs every transfer, and whatever gets trained on Forge sessions ships back out publicly, so the builders who fed it can download and run it themselves later. DeepSeek is making its 75 percent API discount permanent https://startupfortune.com/deepseek-is-making-its-75-percent-api-discount-permanent/ DeepSeek has made its 75 percent V4-Pro API discount permanent after the promotion period, resetting the cost expectations for AI startups and developers. The move puts pricing pressure on OpenAI, Anthropic and Google while raising fresh questions about DeepSeek’s margin strategy as it pursues a major funding round. - deepseek api pricing discount permanent https://startupfortune.com/deepseek-is-making-its-75-percent-api-discount-permanent/ - cheapest frontier ai model for startups https://startupfortune.com/deepseek-is-making-its-75-percent-api-discount-permanent/ That's the deal in plain terms: cheap compute now, in exchange for the raw material Bolt and Arcee need to build something neither of them could train alone. Open-weight models have been closing the gap on proprietary ones for a while now. DeepSeek's R1 release in January 2025 rattled markets specifically because a Chinese lab had matched frontier performance at a fraction of the training cost, and the open-weight field hasn't slowed down since. GLM, out of Zhipu AI, and Kimi, out of Moonshot AI, have followed the same playbook: build something close to frontier, release the weights, let developers do the rest. Bolt's Build Index score is a small, self-reported data point, but it lands right in that trend line. Frankly, giving away 50x usage for a month is cheap advertising if it works. Bolt gets usage data and training material. Developers get a free upgrade. Arcee gets the compute-heavy part of model training, real user sessions, subsidized by someone else's product. A $9 tier for people who were never going to pay full price Bolt also opened sign-ups for a new plan called Bolt Lite, priced at $9 a month and built specifically on top of Forge's open models. Access started rolling out on September 14, distributed through a waitlist that emails out codes as spots open. Sign-ups close October 14, the same day Forge's free usage boost ends. But once someone is in, Bolt says the $9 price holds for as long as they keep the plan. That's the hook. That price point targets students and side-project builders who were never going to pay for a full Pro seat. It also locks in a user base on Forge's open models before the free usage window closes. That's retention, plain and simple. It's exactly the kind of move you'd expect from a company that just spent a month subsidizing usage to collect training data. None of this changes what a $20 or $30 monthly Pro plan gets you on Bolt's closed models. Bolt now has a foot in both camps. One tier runs its best proprietary system. The other, cheap or free, runs open models that, by Bolt's own measurement, already do 91% of the job. The research preview runs one month. What happens after October 14 is the part nobody's promised yet. Whether the discount usage disappears, whether Bolt Lite's $9 price survives contact with real infrastructure costs, and whether Arcee's trillion-parameter model actually ships - none of that is settled. DeepSeek V4 shows how cheaper AI may come from lower precision https://startupfortune.com/deepseek-v4-shows-how-cheaper-ai-may-come-from-lower-precision/ DeepSeek V4 puts FP4 expert weights, stability engineering, and long-context efficiency at the center of the open-model race. For AI founders, the release is a reminder that cheaper intelligence may come from precision and memory design as much as bigger clusters. - how to run cheaper large language models efficiently https://startupfortune.com/deepseek-v4-shows-how-cheaper-ai-may-come-from-lower-precision/ - low precision AI models for startup cost reduction https://startupfortune.com/deepseek-v4-shows-how-cheaper-ai-may-come-from-lower-precision/ Also read: Travis Kalanick's Atoms Hires Vikas Chandra From Meta's Smart Glasses Team https://startupfortune.com/travis-kalanicks-atoms-hires-vikas-chandra-from-metas-smart-glasses-team/ • Oracle Guts Its Workforce With 6am Layoff Emails to Fund Its AI Buildout https://startupfortune.com/oracle-guts-its-workforce-with-6am-layoff-emails-to-fund-its-ai-buildout/ • South Korea Is on Track to Hit One Trillion Dollars in Exports on AI Chips https://startupfortune.com/south-korea-is-on-track-to-hit-one-trillion-dollars-in-exports-on-ai-chips/ This article is posted in Technology News https://startupfortune.com/category/technology/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.