Moonshot AI has presented Kimi K3.1 Pro, its most capable model in the 3-trillion-parameter class with open weights. According to the company, the model is already available through Kimi Work, Kimi Code and the API, together with the K3.1 Pro Highspeed variant, which the company says runs two to three times faster.
The model has 2.8 trillion total parameters, of which 104 billion are active for each token thanks to the Mixture-of-Experts architecture, with a 1-million-token context and multimodal support. Moonshot focuses on three areas: building full game projects for Unreal, Unity and Godot, parametric 3D modeling in Blender and CAD software, and financial work. On the latter, the company claims a 65% reduction in hallucinations, a figure from the company's own modeling work that has not yet been verified by independent tests.
Image: Kimi_Moonshot — https://x.com/Kimi_Moonshot/status/2113680701770068067
In benchmarks published by the company, K3.1 Pro scores 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra (58.2%) and slightly behind Claude Opus 5.5 (66.4%). On BenchCAD, which measures 3D reconstruction from multiple views through CAD code generation, the model reaches 92.7% with tools, approaching Claude Opus 5.5. Moonshot also claims state-of-the-art on OfficeQA Pro v2 with 64.4 points, and gains over K3 of 7.2 points in financial modeling and 11.6 in financial research.
The weights are scheduled for release by November 4, 2026, according to the company's announcement. Those who want to try it now can already use the API on the official Kimi platform, while the company's tech blog collects the full technical details.
No comments yet. Be the first!