IT Home reported on October 24 that in the early morning of today (24th), Moon’s Dark Side Kimi announced the official launch of the first model in the Kimi K3.1 series — K3.1 Pro. It is also the company’s most powerful 3 trillion-level parameter model to date, supporting 1 million contexts.
High-speed model Kimi K3.1 Pro Highspeed has been released. It offers the same intelligence, with an output speed approximately 2 to 3 times higher.
Starting from today, users can use the Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed models through kimi.com, the Kimi App, Kimi Work, Kimi Code, and Kimi API (platform.kimi.com). The complete model weights will be released before November 4.
According to the official introduction by IT Home, on the benchmarks for code, Agent, and knowledge, K3.1 Pro achieves or approaches the level of the strongest closed-source model. Terminal-Bench 4.0 evaluates the capabilities of programming agents through terminal tasks with limited resources in a real environment, with strict specialized tests for each task. K3.1 Pro scored 65.7% in this assessment, surpassing GPT-6 Astra (58.2%), and only having a slight gap compared to Claude Opus 5.5 (66.4%).
Kimi Code Bench 3.0 Long-horizon is an internal benchmark that covers various engineering fields. It focuses on evaluating the multi-step execution, tool invocation, and end-to-end problem-solving capabilities of programming agents. In this test, K3.1 Pro achieved a score of 61.2%, which is higher than K3’s 55.95%.
K3.1 Pro can skillfully use game engines such as Unity, Unreal Engine, and Godot, and combine online research with 3D generation to create complete experiences. The front-end pages, Three.js interactive scenes, and 3D assets it produces show significant improvements in composition, typography, animations, and detail polishing.
The Kimi K3.1 Pro model is skilled in 3D creation and spatial reasoning. In the BenchCAD benchmark tests, K3.1 Pro achieved a tool score of 92.7%, which is close to 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra. The benchmark assesses the ability to reconstruct 3D models from CAD code by generating renderings from multiple perspectives.
According to the introduction, providing a text description, a sketch, or a reference image for K3.1 Pro, it can infer the shape, proportions, and spatial relationships of the object, and reconstruct it into a usable 3D model. It also has a stronger understanding of existing material libraries and model libraries.
K3.1 Pro is proficient in parameterized modeling, such as mechanical parts, architectural models, multi-parameter configurable products, and custom furniture. It does not rely on MCP connectors; it can directly write and run scripts for professional modeling tools like Blender and CAD, complete material construction and refinement, modify individual components, and adjust geometry, materials, and scene layout.
kimi.com and the Kimi app have synchronized the 3D Make feature, allowing online preview and editing of 3D models. The completed models can be directly downloaded for 3D printing, and videos or real-life photos can also be exported to share on social platforms.
Kimi K3.1 Pro can also deliver a complete financial workflow end-to-end. Based on external benchmarks, K3.1 Pro achieved a score of 64.4 in the OfficeQA Pro v2 benchmark test, achieving the current best result (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 has also improved further compared to K3. This benchmark test measures the model’s ability to support the entire investment workflow, covering everything from target screening to profit analysis.
On 3 internal financial benchmarks, K3.1 Pro also fully exceeds K3:
Financial Modelling Benchmark: Examines the formula interactions, financial logic, and final workbook quality in real modeling tasks. K3.1 Pro improves by 14% compared to K3.
Financial Research Benchmark: Used to test whether a model can perform financial research in real-world scenarios, and produces verifiable results under evidence and deadline constraints. K3.1 is 20% higher than K3.
Business Operation Benchmark: In a simulated business environment, it assesses long-term management and decision-making abilities. The K3.1 Pro score is approximately 4.2 times that of K3.
In the internal pressure test focused specifically on factual accuracy, Kimi Hallu Bench Hard, the error rate of individual fact statements for the new model dropped from 23.4% in K3 to 8.3%; the proportion of answers containing at least one factual error was only about 42.4% compared to the previous version. (This is a pressure test on factual accuracy; the reported error rate focuses deliberately on situations where the model is prone to mistakes, and the reported error rate does not represent the overall level in daily use.)
Accessing Kimi API for application development: Visit the Kimi open platform (platform.kimi.com), select the model kimi-k3.1-pro, with a price of 1.3 yuan per million tokens for hit cache and 13 yuan for no hit cache, and an output of 68 yuan. The high-speed version of the model kimi-k3.1-pro-highspeed has an output speed approximately 2 to 3 times faster, at a price 2 times higher.