Moonshot AI releases the open-weight MoE model Kimi K3.1 Pro, and launches a 2 to 3 times faster…

On October 23, 2026, China's Moonshot AI announced the latest version of its multimodal MoE model, "Kimi K3.1 Pro," along with the high-response model "Kimi K3.1 Pro Highspeed."

Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence -kimi.ai

Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.

🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal. 🔹 Full-project coding: games in Unreal, Unity & Godot. 🔹 Excels at parametric 3D modeling in Blender & CAD. 🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%. 🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.

Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on https://t.co/85jA71X9gw, Kimi Work, Kimi Code, and the Kimi API.

Open weights by Nov 4, 2026.

🔗 API: https://t.co/XCrgjXAqMw 🔗 Tech blog: https://t.co/5iPP0MBej6— Kimi.ai (@Kimi_Moonshot) October 23, 2026

Kimi K3.1 Pro is a multimodal MoE (Mixture of Experts) model with a context length of 1 million tokens, 2.8 trillion total parameters, and 104 billion active parameters. Kimi K3.1 Pro Highspeed is its high-speed version, achieving 2 to 3 times the speed of Kimi K3.1 Pro and offering a more responsive user experience.

Kimi K3.1 Pro is a version of the base model Kimi K3 in which the entire reinforcement learning (RL) pipeline has been overhauled; as a result of optimizing every layer related to reinforcement learning, including the RL framework, its compute capability after post-training has been greatly improved. This enables Kimi K3.1 Pro to demonstrate high performance on various benchmarks such as coding, agents, and knowledge. For example, on Terminal-Bench 4.0, which evaluates coding agents, Kimi K3.1 Pro surpassed GPT-6 Astra and came in slightly below Claude Opus 5.5.

Terminal-Bench 4.0 results

In addition, Kimi K3.1 Pro can integrate with game engines such as Unity, Unreal Engine, and Godot, and by combining with web search and 3D generation capabilities, it can create high-quality 3D assets and animations for frontends and interactive scenes. It also excels at 3D creation and spatial reasoning, and can infer shape, proportions, and spatial relationships from textual descriptions, sketches, and reference images to create practical 3D models. Even without MCP connectors, it can create and execute scripts for tools such as Blender and CAD software, building and refining assets, editing components, and adjusting geometry, materials, and scene layouts. In the company's blog, Kimi K3.1 Pro is described as "the most capable open-weight model in the company's history."

In finance as well, Kimi K3.1 Pro achieves highly accurate modeling with a 65% reduction in hallucinations, and on OfficeQA Pro v2, a benchmark that evaluates financial data analysis through search, cross-document verification, and multi-step numerical reasoning, it marks state-of-the-art (SOTA) results on par with GPT-6 Astra and Claude Opus 5.5.

Regarding the recently problematic misuse of AI for cybersecurity capabilities, the company aims to "not release open weights that would expand the absolute limits of cyber capabilities"; specifically, it analyzes and evaluates Kimi K3.1 Pro's vulnerability research capabilities and, during training and the reinforcement learning stage, identifies and removes cyber-related data while retaining the general coding tasks necessary for the model's usefulness. This is said to reduce potential misuse risks while maintaining general usefulness.

Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available on Kimi.ai, the desktop app Kimi Work, and the coding agent Kimi Code, and can also be used via API. The open-weight version is scheduled to be released by November 4, 2026.