Moonshot AI released Kimi K3.1 Pro, a post-trained version of its Kimi K3 flagship, along with a Highspeed variant it rates at two to three times the speed. K3.1 Pro is the first model in a new K3.1 series, and Moonshot describes it as its most capable 3T-class open-weight model so far. The full weights are due by Nov 4, 2026.
Per Moonshot AI, Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0. That puts it ahead of GPT-6 Astra at 58.2% and narrowly behind Claude Opus 5.5 at 66.4%. Moonshot ran its own model through the Claude-Code harness, while each rival's score is the highest one that rival has published or the official leaderboard lists.
Kimi K3.1 Pro keeps the 1-million-token context window of Kimi K3, a 2.8-trillion-parameter model, and adds substantially larger post-training compute, used for reinforcement learning across hundreds of task domains. Moonshot says the gains show up in every domain. On its internal Kimi Code Bench 3.0 Long-horizon test of multi-step coding agents, the score rose from K3's 55.95% to 61.2%.
Moonshot also reports a drop in invented facts. On Kimi Hallu Bench Hard, an internal stress test with examples built to trip models up, the claim-level hallucination rate fell from 23.4% for K3 to 8.3%. The company adds that the test is deliberately adversarial and the rates do not describe everyday error levels.
Both models are live on Kimi.com, Kimi Work, Kimi Code and the Kimi API, where they go by kimi-k3.1-pro and kimi-k3.1-pro-highspeed, and K3.1 Pro is also in the Kimi mobile apps for iOS, Android and HarmonyOS. Kimi Work needs desktop version 3.1.0 or later on Windows or Apple silicon Macs, and Kimi Enterprise administrators have to switch the model on themselves, since access is off by default at launch.
Outside Moonshot's own apps, Vercel has added both models to its AI Gateway as moonshotai/kimi-k3.1-pro and moonshotai/kimi-k3.1-pro-highspeed. Vercel says the gateway charges provider pricing with no markup and that support from more providers will follow.
For the open-weight release, Moonshot says it removed cyber-related data from both the mid-training corpus and the reinforcement-learning mix. On its in-house xdev-bench-hard exploit-development test, the final model solves 81.0% of problems at pass@3, against 90.5% for a checkpoint trained without those removals. The company says K3.1 Pro's offensive cyber ability stays behind proprietary frontier models that have been public for at least three months, and that it intends to raise the cyber capability available in open weights step by step.
