Moonshot AI Releases Kimi K3.1 Pro: 3 Trillion Parameters, 1 Million Context, Coding Agent Benchmarks Surpass GPT-6 Astra

Kimi K3.1 Pro Officially Released: Moonshot AI's Most Powerful 3-Trillion-Parameter Model, Outperforms GPT-6 in Coding Agent Benchmarks

English text is being prepared.

English text is being prepared.

English text is being prepared.

The high-speed variant, Kimi K3.1 Pro Highspeed, is launching at the same time, offering the same intelligence with roughly 2–3x the output speed.

Starting today, users can access Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed via kimi.com, the Kimi App, Kimi Work, Kimi Code, and the Kimi API (platform.kimi.com). Full model weights will be released by November 4.

English text is being prepared.

Kimi Code Bench 3.0 Long-horizon is an internal benchmark covering multiple engineering domains, focusing on coding agents' multi-step execution, tool use, and end-to-end problem-solving capabilities. K3.1 Pro scored 61.2% on this test, up from K3's 55.95%.

K3.1 Pro is proficient with game engines such as Unity, Unreal Engine, and Godot, and can combine web search with 3D generation to create complete experiences. The front-end pages, Three.js interactive scenes, and 3D assets it produces show clear improvements in composition, layout, motion, and detail refinement.

The Kimi K3.1 Pro model excels at 3D creation and spatial reasoning. On the BenchCAD benchmark, K3.1 Pro scored 92.7% with tool use, close to Claude Opus 5.5's 96.2% and GPT-6 Astra's 95.9%. This benchmark assesses the ability to reconstruct 3D models from multi-view rendered images by generating CAD code.

According to the introduction, given a text description, a sketch, or a reference image, K3.1 Pro can infer an object's shape, proportions, and spatial relationships and reconstruct it as a usable 3D model. It also has a stronger understanding of existing asset and model libraries.

K3.1 Pro excels at parametric modeling, such as mechanical parts, architectural models, multi-parameter configurable products, and custom furniture. Without relying on MCP connectors, it can directly write and run scripts for professional modeling tools such as Blender and CAD to build and refine assets, modify individual components, and adjust geometry, materials, and scene layout.

kimi.com and the Kimi app have also launched the 3D Make feature, supporting online preview and editing of 3D models; finished models can be downloaded directly for 3D printing, or exported as videos or Live Photos to share on social platforms.

Kimi K3.1 Pro can also deliver complete financial workflows end to end. On external benchmarks, K3.1 Pro scored 64.4 on the OfficeQA Pro v2 benchmark, achieving the current state of the art (SoTA). OfficeQA Pro v2 assesses complex financial data analysis based on real financial documents, involving retrieval, cross-document verification, and multi-step numerical reasoning.

K3.1 Pro's performance on Frontier Finance also improved further over K3. This benchmark measures the model's ability to support complete investment workflows, covering the entire process from target screening to profitability analysis.

On 3 internal financial benchmarks, K3.1 Pro also comprehensively surpasses K3:

Financial Modelling Benchmark: Financial Modelling examines formula linking, financial logic, and final workbook quality in real modelling tasks; K3.1 Pro improves by 14% over K3.

Financial Research Benchmark: Financial Research is used to test whether the model can complete financial research work in real-world scenarios and output verifiable results under the constraints of evidence and deadlines. K3.1 improves by 20% over K3.

Business Operation Benchmark: Business Operation examines long-horizon business management and decision-making capabilities in a simulated business environment; K3.1 Pro scores approximately 4.2 times that of K3.

In the internal stress test Kimi Hallu Bench Hard, conducted specifically for factual accuracy, the new model's error rate on individual factual statements dropped from 23.4% for K3 to 8.3%; the proportion of responses containing at least one factual error was only about 42.4% of that of the previous generation. (This is a stress test for factual accuracy; the samples deliberately focus on situations where the model is prone to errors, and the reported error rates do not represent the overall level in everyday use.)

Developing applications via the Kimi API: visit the Kimi Open Platform (platform.kimi.com) and select the model kimi-k3.1-pro. The price is 1.3 yuan per million tokens for input (cache hit) and 13 yuan (cache miss), and 68 yuan for output. The high-speed model kimi-k3.1-pro-highspeed offers output speed about 2–3 times faster at 2 times the price.