Businesses gain access to Kimi K3.1 Pro, a new open-weight model with a 1-million-token context window, stronger coding and financial research capabilities, and improved factual accuracy. Kimi K3.1 Pro Highspeed delivers two to three times the speed of the standard model, while both versions are available through Kimi products and the API.
Kimi also plans to release the full model weights by Nov. 4, 2026. The release is planned to give developers and organizations access to a model intended for long-running coding, 3D creation, financial analysis, and other agentic tasks.
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is the first model in the K3.1 series. It is a post-trained version of Kimi K3 and is described by Kimi as its most capable 3T-class open-weight model to date.
The company used substantially more post-training compute for the release and optimized the reinforcement learning framework, long-context inference, training efficiency, sandbox infrastructure, rollout scheduling, and environment quality checks. Kimi says these changes helped it conduct joint reinforcement learning across hundreds of downstream domains.
Its 1-million-token context window can accommodate long codebases, extensive financial documents and multi-stage research tasks. Kimi K3.1 Pro Highspeed is designed for faster interactions and is available alongside the standard version.
How does Kimi K3.1 Pro perform?
According to Kimi’s reported benchmark results, K3.1 Pro scored 65.7% on Terminal-Bench 4.0. That was ahead of GPT-6 Astra at 58.2% and slightly below Claude Opus 5.5 at 66.4%.
On Kimi Code Bench 3.0 Long-horizon, an internal benchmark covering multi-step execution, tool use and end-to-end engineering work, K3.1 Pro scored 61.2%. Kimi K3 received 55.95%.
The model also posted 92.7% with tools on BenchCAD, a benchmark for reconstructing 3D objects from multiple views through CAD code generation. That result was below Claude Opus 5.5 at 96.2% and GPT-6 Astra at 95.9%.
For financial work, K3.1 Pro achieved 64.4 on OfficeQA Pro v2, which tests complex fiscal analysis involving real financial documents, retrieval, cross-document verification and multi-step numerical reasoning.
Against Kimi K3, the new model improved by 7.2 points on Financial Modeling and 11.6 points on Financial Research. On Business Operation, a benchmark for long-horizon management and decision-making in a simulated business environment, K3.1 Pro achieved about four times K3’s performance.
What improved for coding and 3D creation?
Kimi says K3.1 Pro can handle quick code fixes, complete project builds, games and interactive web experiences. It works with Unity, Unreal Engine and Godot, and can combine those tools with web search and 3D generation.
The release also includes improvements to front-end pages, Three.js scenes and 3D assets, with particular attention to composition, typography, animation and visual polish.
In an internal walking-dynamics test, K3.1 Pro used the open-source human-atlas model, which contains 296 skeletal components and 402 muscle components. Across 167 missions and 420 independent reviews, the system analyzed results, identified problems and refined its controller. The simulated human progressed from collapsing within seconds to walking continuously with alternating strides for 11.2 seconds.
When reward adjustments did not improve arm motion, the model ran open-loop probes, corrected 36 errors in shoulder-muscle moment-arm calculations and retrained the controller.
For 3D work, K3.1 Pro can create models from a text brief, sketch or reference image. It also supports parametric modeling for mechanical parts, architectural designs, configurable products and custom furniture. Without MCP connectors, it can write and run scripts for Blender and CAD software, then refine geometry, materials, components and scene layouts.
3D Make, available on kimi.ai and in the Kimi app, lets users preview and edit models, download files for 3D printing, and export models as videos or Live Photos.
How does the model handle factual accuracy?
Kimi made targeted changes to improve factual accuracy on real-world tasks and difficult prompts. On the internal Kimi Hallu Bench Hard benchmark, its claim-level hallucination rate fell from 23.4% for K3 to 8.3%.
The share of responses containing at least one factual error was 42.4% of K3’s corresponding rate. Kimi cautions that the benchmark is a stress test containing deliberately difficult examples, so the figures do not represent overall error rates in daily use.
What controls apply to cybersecurity capabilities?
Before releasing K3.1 Pro’s weights, Kimi removed cyber-related data from mid-training and filtered cyber-related tasks from its reinforcement learning environment. The company retained general-purpose coding tasks needed for practical usefulness.
On the internal xdev-bench-hard benchmark, a checkpoint without those controls solved 90.5% of tasks at pass@3. The final K3.1 Pro reached 81.0%. Kimi says the difference shows that the data removal meaningfully reduced cyber capability relative to the same training pipeline without the controls.
The benchmark uses source-code-based vulnerability discovery and exploitation tasks drawn from disclosed issues in software such as the Linux kernel, Node.js, JavaScriptCore, Exim and VirGL. Kimi also says K3.1 Pro remains behind frontier proprietary models in offensive cyber capability.
Where is Kimi K3.1 Pro available?
Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available through Kimi.ai, the Kimi mobile apps, Kimi Work, Kimi Code and the Kimi API. Full model weights are scheduled for release by Nov. 4, 2026.
- Kimi mobile apps are available for iOS, Android and HarmonyOS.
- Kimi Work version 3.1.0 or later is available for Windows and Apple silicon Macs.
- Kimi Code users can select Kimi K3.1 Pro with the /model command.
- API users can select kimi-k3.1-pro or kimi-k3.1-pro-highspeed.
- Zero Data Retention is available for API customers.
- Kimi Enterprise administrators can enable K3.1 Pro for their organization, with access off by default at launch.
What to do now
- Review whether long codebases, extensive financial documents or multi-stage agentic work could benefit from a 1-million-token context window.
- Test K3.1 Pro in the relevant Kimi product or API environment against a controlled set of current business workflows.
- Evaluate K3.1 Pro Highspeed when response time matters more than the standard model’s processing speed.
- Plan for the Nov. 4, 2026 full-weight release if operating the model on infrastructure you control is important to your organization.
- Review factual outputs and source grounding in finance, research and other high-consequence workflows before incorporating model results into business decisions.
FAQ
What is the context window of Kimi K3.1 Pro?
Kimi K3.1 Pro has a 1-million-token context window.
How fast is Kimi K3.1 Pro Highspeed?
Kimi says Kimi K3.1 Pro Highspeed delivers two to three times the speed of Kimi K3.1 Pro.
When will the full Kimi K3.1 Pro model weights be released?
The full model weights are scheduled for release by Nov. 4, 2026.
Related coverage
- Tech Firms Urge Washington Not to Restrict Open-Weight AI
- Poolside Laguna S 2.1 open-weight coding model released
- Moonshot AI’s Kimi K3 Beats Opus 4.8 as the Largest Open Model Ever – BizScoreAI
Sources
See our editorial disclaimer for how our articles are produced.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.