Skip to content

An editorial team and a community building with AI

vibecoding.incJoin Discord ↗
← All articles

News · Moonshot AI (Kimi) · 2 min read

Event date: · Published:

Moonshot releases Kimi K3.1 Pro with 1M context and promises open weights

On October 24, 2026, Moonshot AI released Kimi K3.1 Pro, a post-trained Kimi K3 with a 1M context, in Kimi apps, Kimi Work, Kimi Code and the API, and said full model weights would follow by November 4.

Archive entry: this covers an event from October 24, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 24, 2026, Moonshot AI released Kimi K3.1 Pro, a post-trained version of Kimi K3 with larger post-training compute and a 1M context, as reported on the Kimi research and tech blog. It is the first model of the K3.1 series, and Moonshot described it as the most capable 3T-class open-weight model to date.

What changed

  • A Kimi K3.1 Pro Highspeed variant runs 2-3x faster.
  • It was available on Kimi.ai, the iOS, Android and HarmonyOS apps, Kimi Work 3.1.0+ for Windows and Apple silicon, and Kimi Code via /model.
  • The API models are kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Zero Data Retention is available, and enterprise access is off by default.
  • Moonshot reported 65.7% on Terminal-Bench 4.0, versus 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5, and 61.2% on Kimi Code Bench 3.0 long-horizon, versus 55.95% for K3.
  • The hallucination rate fell from 23.4% to 8.3%. Moonshot removed cyber data as a mitigation, and xdev-bench-hard is 81.0%, versus 90.5% without that step.
  • API pricing is flat pay-as-you-go with no context tiering. Caching is automatic with a 5-minute or 1-hour TTL, batch costs 60% of the standard price, and the API unlocks after a minimum $1 top-up.

What it means for people building with AI

Kimi Code users and API developers gained access to a model that Moonshot's own benchmarks place close to Claude Opus 5.5 on Terminal-Bench 4.0, 65.7% versus 66.4%. Self-hosters were promised the weights under Moonshot's own license.

Open weights and further reading

Moonshot said full model weights would be released by November 4, 2026. A Hugging Face repository was created on November 2, 2026, and its model card describes a mixture-of-experts model with 2.8T total and 104B activated parameters, 93 layers and 896 experts (16 selected per token, 2 shared), with MXFP4 weights.

Details are in the Kimi K3.1 Pro tech blog, the API quickstart and the pricing page.

Sources

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.

Moonshot releases Kimi K3.1 Pro with 1M context and promises open weights | vibecoding.inc