Kimi K3.1 Pro Launches With Advanced Coding and 3D Modeling Capabilities
Moonshot AI has launched Kimi K3.1 Pro, the first model in its new K3.1 series, with improvements in software development, 3D modeling, financial research, and complex AI agent workflows. The model features a context window of up to 1 million tokens, allowing it to process large amounts of information when working on demanding, multi-step tasks.
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… pic.twitter.com/klydcszv02– Kimi.ai (@Kimi_Moonshot) October 23, 2026
The launch also includes Kimi K3.1 Pro Highspeed, a variant that the developer notes delivers two to three times the speed of the standard model. Both versions are available through Kimi’s website and product ecosystem, while the company plans to release the model’s full weights by November 4, 2026.
Kimi K3.1 Pro Improves Coding and AI Agent Workflows
Kimi K3.1 Pro is designed to handle more than short code-generation requests. It targets longer development tasks involving debugging, tool use, project construction, and repeated testing until a solution works as intended.
According to Moonshot AI, the model can build complete web experiences, interactive Three.js scenes, and game projects using tools suited to each task. It supports workflows involving game engines such as Unity, Unreal Engine, and Godot, combining coding with web search and 3D asset generation when needed.
The company reports a score of 65.7% on Terminal-Bench 4.0, a benchmark that evaluates coding agents on practical command-line tasks. That result places Kimi K3.1 Pro ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%, based on the figures cited in the launch announcement.
On Moonshot AI’s internal Kimi Code Bench 3.0 Long-horizon evaluation, which tests multi-step execution, tool use, and end-to-end engineering problem-solving, the new model scored 61.2%, compared with 55.95% for Kimi K3.
The company also highlighted a more complex experiment involving a simulated human walking system. Across 167 missions and 420 independent reviews, Kimi K3.1 Pro analyzed test results, diagnosed issues, and refined the controller. It identified and corrected 36 errors in shoulder-muscle calculations, eventually enabling the simulated character to walk with alternating strides for 11.2 seconds. The example illustrates the model’s ability to investigate failed results and revise its approach rather than simply generate code once.
3D Modeling From Text, Sketches, and Images
3D creation is another major focus of Kimi K3.1 Pro. The model can use a text description, sketch, or reference image to infer shapes, proportions, and spatial relationships before generating a usable 3D model.
In BenchCAD, an evaluation that measures 3D reconstruction through CAD code generation, Kimi K3.1 Pro scored 92.7% when using tools. Moonshot AI reported scores of 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra on the same evaluation.
The model is also designed for parametric modeling, where dimensions and relationships can be adjusted without rebuilding an entire object. Potential applications include mechanical components, architectural models, configurable products, and custom furniture. It can write and run scripts for tools such as Blender and CAD software to refine geometry, materials, individual components, and scene layouts.
Alongside the model launch, Moonshot AI made its 3D Make feature available through Kimi’s website and app. Users can preview and edit generated models online, download files for 3D printing, or export their creations as videos and Live Photos. This provides a more direct route from an initial idea to a visual asset that can be shared or used in a project.
Financial Research and Factual Accuracy
Kimi K3.1 Pro also targets financial workflows that require retrieving information from documents, checking figures across sources, performing calculations, and connecting evidence to conclusions.
On OfficeQA Pro v2, a benchmark focused on complex financial analysis using real-world documents, the model scored 64.4, which Moonshot AI described as a leading result. The company also reported improvements over Kimi K3 on Frontier Finance, which evaluates broader investment workflows, from opportunity screening to profitability analysis.
Its internal financial evaluations showed gains in several areas. Kimi K3.1 Pro improved by 7.2 points on financial modeling and 11.6 points on financial research compared with Kimi K3. On a simulated business operations benchmark involving long-term management and decision-making, it achieved approximately 4.2 times the predecessor’s performance.
Moonshot AI also reported a lower hallucination rate on its internal Kimi Hallu Bench Hard evaluation. The claim-level rate fell from 23.4% for Kimi K3 to 8.3% for Kimi K3.1 Pro. However, the company notes that this benchmark deliberately uses difficult prompts designed to expose factual errors, so these figures should not be interpreted as everyday error rates.
These results suggest a stronger focus on evidence-based analysis, although independent evaluations and real-world use will be important for determining how consistently the improvements hold up outside the company’s testing environment.
Cybersecurity Safeguards for the Open-Weight Release
Moonshot AI also outlined measures intended to reduce the risk of misuse as it prepares to release Kimi K3.1 Pro’s full model weights.
According to the company, it identified and removed cybersecurity-related data from specific stages of training while retaining general-purpose coding capabilities. On its internal xdev-bench-hard evaluation, which tests vulnerability discovery and exploitation, the final model scored 81%, compared with 90.5% for an experimental checkpoint without those risk-reduction measures.
The developers explain that the approach is intended to preserve useful capabilities for software development and defensive security while limiting potential misuse. The planned open-weight release will give developers more flexibility to run and adapt the model in their own environments, while also making responsible deployment an important consideration.
How to Access Kimi K3.1 Pro
Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available through Kimi’s official website, its mobile apps, Kimi Work, and Kimi Code. Developers can also access the models through the Kimi API platform, using the model identifiers kimi-k3.1-pro and kimi-k3.1-pro-highspeed.
In Kimi Code, users can select the model with the /model command. The full model weights are scheduled for release by November 4, 2026, according to Moonshot AI’s announcement.
Kimi K3.1 Pro expands Moonshot AI’s focus on models that can complete extended workflows involving reasoning, coding, and external tools. Its reported improvements in software development, 3D creation, and financial research make it a notable release, but independent benchmarks and hands-on testing will help establish how its capabilities compare with competing models in everyday use.
