Kimi K3.1 Pro jumps four spots

Moonshot AI has launched Kimi K3.1 Pro, a new reasoning model that quickly climbed to the fourth position on the Artificial Analysis Intelligence Index. The system achieved a score of 55 points, marking an 11 point increase over its predecessor, Kimi K3. This performance places it just three points behind the leading proprietary systems currently available in the market.

The new model is a post-trained version of K3 that utilized significantly more computational resources during its training phase. According to data from Artificial Analysis, only three models currently outperform it. Anthropic's Claude Opus 5.5 leads with a score of 58. Claude Sonnet 5.5 and Muse Wonder Full both sit at 56 points. Moonshot intends to publish the model weights by November 4, 2026. If this release happens as planned, K3.1 Pro would become the highest scoring open-weights model available. This move would narrow the gap between open and proprietary models from 12 points to just three.

Alongside the standard version, the company offers K3.1 Pro Highspeed. This variant is designed to run two to three times faster than the base model. Both versions are accessible through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. This wide availability allows users to test the model across various professional and developer environments immediately.

The most significant performance gains appeared in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached an Elo rating of 1,846. This is 312 points higher than K3 and sits just behind Claude Opus 5.5 and Muse Wonder Full. In coding tests, the model showed strong progress. On Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent. This places it ahead of GPT-6 Astra and slightly below Claude Opus 5.5. By comparison, the older K3 model managed only 13 percent on that specific test.

The company also reports a major reduction in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent with K3 to 8.3 percent. Moonshot describes this benchmark as a stress test rather than a measure of general error rates. GPT-6 Astra still leads this category with a rate of 5.5 percent. Financial benchmarks also showed improvement. The model gained points in Financial Modeling and Financial Research. Moonshot claims it tops all tested models on OfficeQA Pro v2.

Capabilities in 3D modeling and coding were also highlighted. K3.1 Pro works with game engines such as Unity, Unreal Engine, and Godot. It can write and run scripts for Blender and CAD software without needing MCP connectors. On the BenchCAD benchmark, it scores 92.7 percent with tools available. This is slightly lower than Claude Opus 5.5 and GPT-6 Astra. Without tools, the model reaches 72.8 percent. Moonshot has added a 3D Make feature on Kimi.com to allow users to preview, edit, and export 3D models directly.

The underlying architecture remains the same as K3. It features 2.8 trillion total parameters, with 104 billion active. The model supports a one million token context window and handles both text and images. Artificial Analysis places it at the top of the Pareto frontier for its size. API prices have dropped by 33 percent. Input costs are now $2 per million tokens, output is $10, and cache hits cost $0.20. Despite generating 14 percent more output tokens, the cost per task remains nearly identical to the previous version.

Moonshot addressed cybersecurity concerns in its technical documentation. The company noted that vulnerability research capabilities improved as post-training scaled. Cyber-related content was part of the general mid-training corpus. For the open-weights release, this data was removed from both training phases. On an internal benchmark that reproduces recent vulnerabilities in the Linux kernel and Node.js, the uncleaned model solved 90.5 percent of tasks. After cleaning, that number dropped to 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent on the same test.

The company believes that no open-weights model should push the absolute frontier of offensive cyber capabilities. However, it aims to raise what open models can do gradually to support defensive work like auditing legacy systems. Anthropic previously demonstrated that even models with lower intelligence scores can write exploits at high levels. This suggests that the balance between open access and security remains a key discussion point in the AI industry as these models continue to evolve.