Kimi K3.1 Pro - Moonshot’s 3T-class model for coding, 3D and finance
by•
Kimi K3.1 Pro is Moonshot’s most capable 3T-class model, a post-trained update to K3 with a 1M-token context. Moonshot reports gains in coding, 3D modeling, finance and agentic tasks, including 65.7% on Terminal-Bench 4.0. A Highspeed version runs 2-3x faster. It’s live on Kimi.ai, the apps, Kimi Code and the API. Moonshot says open weights are due by November 4, 2026.
Replies
Kimi K3.1 Pro is Moonshot’s new model, the first in the K3.1 series. It’s a post-trained update to K3, with substantially more post-training compute and a 1M-token context.
For context, K3 is Moonshot’s 3T-class model. Moonshot says K3.1 Pro is its most capable open-weight model at that scale, with full weights due by November 4, 2026.
What changed in this update:
More post-training compute, and RL infrastructure optimized for joint RL across hundreds of domains
Better coding: it works with Unity, Unreal Engine and Godot, and builds front-end pages and Three.js scenes, with stronger long-horizon coding
3D modeling: it writes and runs scripts for tools like Blender and CAD software, and Moonshot launched 3D Make on kimi.ai and the Kimi app, with export for 3D printing
Finance: aimed at end-to-end workflows like modeling, research and business-operation simulation
Better factual accuracy and agentic performance
Kimi K3.1 Pro Highspeed, which Moonshot says runs 2-3x faster
Some numbers:
Terminal-Bench 4.0: 65.7%, against GPT-6 Astra at 58.2% and Claude Opus 5.5 at 66.4%
Kimi Code Bench 3.0 Long-horizon: 61.2%, up from K3’s 55.95%
OfficeQA Pro v2: 64.4%
Claim-level hallucination on a hard stress test, with tools: 8.3%, down from K3’s 23.4%
It’s live now on kimi.ai, the Kimi apps, Kimi Work, Kimi Code and the Kimi API (kimi-k3.1-pro and kimi-k3.1-pro-highspeed). Kimi Business admins can enable it, and it’s off by default.
Good fit if you build with coding agents, 3D or CAD workflows, or financial analysis and want a model with a long context.