Moonshot AI launches Kimi K3.1 Pro, a 1M-token open-weight model

Moonshot AI released Kimi K3.1 Pro, which it calls its most capable 3T-class open-weight model yet, with a 1-million-token context window. It is a post-trained version of Kimi K3 built with much larger post-training compute, and the company also released a faster variant called K3.1 Pro Highspeed, said to run 2 to 3 times faster.

According to Moonshot AI's own benchmarks, K3.1 Pro scores 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra's 58.2% and just behind Claude Opus 5.5's 66.4%. The company says the model improves at coding, 3D modeling and CAD work, financial research tasks, and game-engine work in Unity, Unreal Engine, and Godot. It also highlights a drop in its internal hallucination benchmark, with the claim-level error rate falling from 23.4% to 8.3% compared to K3.

Moonshot AI says it deliberately removed cyber-related training data before this open-weight release to limit how much offensive security capability the model carries, while keeping general coding ability. The model is live now on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API, with full model weights due by November 4.

Why it matters

It's a major new open-weight model claiming near-frontier coding and reasoning performance, with a notably large context window.

Source: Moonshot AI