Kimi K3 changelog

Kimi K3.1 Pro and K3.1 Pro Highspeed released

Moonshot released Kimi K3.1 Pro, a post-trained version of Kimi K3 with a 1-million-token context window, plus a Highspeed variant with 2-3x the speed. Both are available on Kimi.com, Kimi Work, Kimi Code and the Kimi API, and full open weights are due by Nov 4, 2026.

Who is affected:

Kimi app, Kimi Code and Kimi API users get a stronger coding, 3D and finance model, and Kimi Enterprise admins must enable it because access is off by default.

What to do:

Select kimi-k3.1-pro or kimi-k3.1-pro-highspeed in the Kimi API, or use /model in Kimi Code.

Vals Index: Muse Wonder Full fourth, DeepSeek V4.1 Flash 1017 eighth at $0.66 per test

The Vals Index, updated October 22, 2026, ranks Muse Wonder Full fourth at 66.79% and Muse Wonder sixth at 64.88%. DeepSeek V4.1 Flash 1017 is eighth at 61.26% for $0.66 per test, GLM 5.3 ninth at 59.73%, Grok 4.7 12th at 58.56% and Kimi K3 19th at 52.95%.

Kimi K3 available in GitHub Copilot

Kimi K3, hosted by GitHub on Fireworks AI, rolled out to Copilot Pro, Pro+, Max, Business and Enterprise at $3 per 1M input tokens, $15 per 1M output tokens and $0.30 per 1M cached input tokens.

Who is affected:

Copilot users looking for a cheaper frontier-level coding model.

What to do:

Business and Enterprise admins must enable the Kimi K3 policy.

Kimi K3 full weights published on Hugging Face

Moonshot published the full Kimi K3 weights (2.8T parameters, 1M-token context) on Hugging Face under the Kimi K3 License, as promised at launch.

Who is affected:

Teams that self-host open-weight models.

What to do:

Read the Kimi K3 License before deploying the weights.

Moonshot pauses new Kimi subscriptions after Kimi K3 demand

Moonshot said demand in the 48 hours after the Kimi K3 launch pushed its compute close to capacity, so it suspended new consumer subscriptions with immediate effect. Existing subscribers keep their benefits.

Moonshot releases Kimi K3, full weights promised by July 27

Moonshot introduced Kimi K3, a 2.8T-parameter model with native vision and a 1M-token context window, available on Kimi.com and the Kimi API, with full weights to be released by July 27, 2026.