Bolt opens its open-model tier to everyone, and GLM 5.3 Flash takes the crown

Bolt announced Monday it is opening access to open models for more than 11 million users, with a usage quota up to 50 times larger than before. The more revealing number came from the company's own leaderboard: Z.ai's GLM 5.3 Flash captured 54% of selections, leaving DeepSeek V4 Pro at 17%, the standard GLM 5.3 at 15%, and Kimi K3 at 14%. The winning formula, Bolt said, was a blend of speed, low cost, and a context window twice the size.

Open models go live

The move puts Bolt in territory where its main rivals — Cursor, Windsurf, and GitHub Copilot — still lean heavily on closed models. The practical upshot: developers can now run longer coding tasks without hitting token ceilings, and pay less per run. That a Chinese model (GLM) tops the table by a wide margin signals the gap between Western and open models has narrowed dramatically in recent months.

New pricing: Bolt Lite at $9 a month

Alongside the rollout, Bolt launched Bolt Lite, a $9-per-month plan aimed at students, weekend developers, and builders who want to reach production without paying full freight. The free tier stays a tasting menu; Lite unlocks the full capabilities of the new engine. No waitlist for the next 72 hours; an access code will arrive by email when capacity opens.

Forge: the engine under the hood

The infrastructure enabling all of this is called Bolt Forge, an open-source model mode that serves as the foundation for the Lite plan. Forge launched officially on September 14 and is architected so that any new frontier open model can plug in without integration lag. It's a modular approach reminiscent of what LLM gateways do, only packaged inside Bolt's development environment.

What this means for developers in practice

Bottom line: anyone building web apps in Bolt now gets more context, fewer cost guardrails, and the ability to swap models with a click when something better ships. The catch — still no independent benchmarks verifying the internal ranking, and production performance will vary by task type. But the direction is clear: the model layer is commoditizing, and the value is migrating to the tooling wrapped around it.