Bolt launches Forge with DeepSeek V4.1 Flash and a $9-a-month Lite plan

Bolt has opened Forge, its new open-models mode, and the first week's data shows developers voted with their feet: DeepSeek V4.1 Flash logged ten times more runs than V4 Pro. Beta testers reported that Flash produces the best-looking sites — page structure, design consistency, CSS handling — even though it is nominally the "lighter" variant. Bolt attributes the edge to specific optimization for front-end coding tasks, not general benchmarks.

Forge is a runtime built entirely on open-weight models, designed to ingest each new frontier release as it arrives. The official launch began September 14. The goal is a fast lane that drops the best publicly available models straight into Bolt's development environment, without waiting for licensing deals or closed-vendor approvals.

At the same time, Bolt is rolling out Bolt Lite, a $9-per-month tier aimed at students, weekend developers and early-stage product builders. For the next 72 hours there is no waitlist; after that the invitation system returns. The free tier remains for "kicking the tires" — trying, testing, learning — while Lite provides quotas large enough to finish a project and ship it, using the same builder and interface.

The move positions Bolt as an intermediary layer that feeds the best of the open ecosystem to users, rather than locking them into a single closed model. If the next Qwen or Llama release outperforms DeepSeek on front-end code, it will enter Forge without users needing to switch platforms. For teams operating on a "ship fast, iterate short" cycle, that lowers the cost of experimentation and lets them swap engines without changing vehicles.