Kimi K3.1 Pro is Moonshot AI's newest open-weight language model, announced on the Kimi blog in October 2026. It is a heavily post-trained version of Kimi K3 with a 1-million-token context window, and Moonshot calls it its most capable 3T-class open-weight model so far. A companion model, Kimi K3.1 Pro Highspeed, runs 2–3× faster for more responsive interactions.
What is new in Kimi K3.1 Pro
- Same base, more training: it is Kimi K3 with substantially larger post-training compute, driven by large-scale reinforcement learning across hundreds of domains.
- Availability: the models are in the Kimi apps and the Kimi API now, and Moonshot says the full weights will be released by 4 November.
- Target work: agentic coding, 3D and CAD creation, financial research and other long-horizon agent tasks.
Benchmarks Moonshot reports
All figures below come from Moonshot's own announcement and have not been independently replicated.
- Terminal-Bench 4.0: 65.7%, ahead of GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%).
- BenchCAD, which rebuilds objects as CAD code from multi-view images: 92.7% with tools, behind Opus 5.5 (96.2%) and GPT-6 Astra (95.9%).
- OfficeQA Pro v2, financial analysis over real documents: a state-of-the-art 64.4.
- Internal hallucination stress test: the claim-level error rate fell from 23.4% for K3 to 8.3%. Moonshot notes this is a stress test, not an everyday error rate.
Long-horizon example from the post: over 167 missions, the model kept refining a simulated human until it walked with alternating strides for 11.2 seconds instead of collapsing.
Why Moonshot cut cyber data from the open weights
The most unusual section of the announcement is about safety. Moonshot says it found and removed cyber-related data from both mid-training and reinforcement learning for the open-weight release, while keeping general coding tasks. On its internal exploit benchmark, a version trained without those cuts solved 90.5% of problems; the shipped model solves 81.0%. In other words, its offensive skill is capped on purpose, not removed.
For companies, one practical detail: Kimi Enterprise admins have to switch K3.1 Pro on themselves, because access is off by default at launch.
Should you use it?
If you need frontier-level coding on infrastructure you control, mark 4 November and test the weights on your own repositories before trusting vendor benchmarks. Teams new to agents can start with our guides to agentic design patterns and to securing autonomous agents.
