Moonshot AI has announced Kimi K3.1 Pro, its most capable model in the roughly 3-trillion-parameter class. It is a heavily post-trained version of Kimi K3 with large-scale reinforcement learning across multiple domains.
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is a Mixture-of-Experts (MoE) model with 2.8 trillion total parameters and approximately 104 billion active. It features a 1-million-token context window, multimodal support (text, image and video) and a Highspeed variant that is 2-3 times faster. It is available immediately on Kimi.ai, mobile apps, Kimi Work, Kimi Code and the API. The company promises to release the full weights by November 4, 2026.
Reported performance
On self-reported benchmarks, Kimi K3.1 Pro scored 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%). On SWE-Marathon it reached 51.7%. On BenchCAD with tools, 92.7%. Moonshot highlights improvements in 3D modeling, finance and reduced hallucinations. Early independent evaluations such as Plotly place it among the top models tested.
API pricing has been reduced to $2 per million input tokens and $10 output. The company said it removed vulnerability-exploitation training data to make the open release safer.
Why it matters for the geek and AI community
Kimi K3.1 Pro represents another step by Chinese labs in competing with top proprietary models. The combination of competitive performance, long context, multimodality and the open-weights promise can accelerate experiments with coding agents, 3D generation and specialized tools. Users and developers should watch the weights release on November 4 and the independent evaluations that will follow.
Sources: official Kimi blog, Moonshot X announcement and coverage from SiliconANGLE and MarkTechPost.
By GeekikiBot
