Moonshot AI narrows the gap with top US rivals via new Kimi model

Moonshot AI has released Kimi K3.1 Pro, a new reasoning model that has climbed to fourth place on the Artificial Analysis Intelligence Index. The model scores 55 points, placing it just three points behind Anthropic’s Claude Opus 5.5, which holds the top spot with 58 points. This narrow margin marks a significant shift in the competitive landscape, as Chinese developers continue to close the performance gap with leading American firms.

The journey to this release involved substantial engineering efforts. Moonshot took its previous Kimi K3 architecture and applied significantly more post-training compute. This effort resulted in an 11-point jump in overall intelligence scores. Only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full currently rank higher on the index. If Moonshot follows through on its plan to release the model weights by November 4, 2026, K3.1 Pro will become the highest-scoring open-weights model available. That move would shrink the performance difference between open and proprietary models from 12 points to just three.

The most dramatic improvements are visible in agentic tasks, where the model demonstrates advanced problem-solving abilities. On the GDPval-AA benchmark, K3.1 Pro achieves an Elo rating of 1,846, a gain of 312 points over its predecessor. It now trails only Claude Opus 5.5 and Muse Wonder Full in this category. On Terminal-Bench 4.0, the model reports a success rate of 65.7 percent, surpassing GPT-6 Astra and coming within a fraction of a point of Claude Opus 5.5. By comparison, the older K3 version managed only 13 percent on that same test.

Reliability has also improved markedly. Moonshot reports that claim-level hallucinations have dropped from 23.4 percent in the previous version to 8.3 percent on its internal stress test. While GPT-6 Astra still holds an edge with a 5.5 percent rate, the reduction indicates a more stable system. The model also shows gains in specialized fields, topping all tested models on OfficeQA Pro v2 for general office tasks and seeing notable improvements in financial modeling and research benchmarks.

A key differentiator for this release is the integration of 3D modeling and coding capabilities. The model can interact directly with game engines such as Unity, Unreal Engine, and Godot. It handles both minor fixes and full project builds, scoring 92.7 percent on the BenchCAD benchmark when using tools. This performance is nearly on par with Claude Opus 5.5 and GPT-6 Astra. The company emphasizes its strength in parametric modeling for mechanical and architectural parts, allowing users to write and run scripts for Blender and CAD software without additional connectors.

Despite these capability gains, the underlying architecture remains consistent in size. K3.1 Pro retains the 2.8 trillion total parameters and 104 billion active parameters of the previous model, along with a one million token context window. However, the cost of using the model has decreased. Input prices have dropped to $2 per million tokens, and output prices to $10 per million tokens, representing a 33 percent reduction from the previous pricing structure.

Moonshot has also addressed cybersecurity concerns surrounding the new release. The company observed that vulnerability research capabilities increased as post-training scaled up. To manage this for the open-weights version, they removed cyber-related data from both training phases. This adjustment reduced the model’s ability to solve tasks on an internal benchmark from 90.5 percent to 81.0 percent. While this is still higher than the previous K3 version, it is slightly below DeepSeek-V4.1-Flash. The company argues that open models should not push the absolute frontier of offensive capabilities but should still serve defensive purposes.

The immediate availability of the model through various APIs and interfaces signals a strategic push into the enterprise market. A faster variant, K3.1 Pro Highspeed, is also available and runs two to three times quicker than the standard version. As the release date for the model weights approaches, the industry will watch to see if the theoretical performance gains translate into real-world dominance for open-weights systems.

The entry of Kimi K3.1 Pro into the top tier of the Intelligence Index forces a reevaluation of the cost versus performance tradeoff. With prices dropping and performance rising, the pressure on proprietary providers to maintain a lead is intensifying. The coming months will reveal whether the open-weights release solidifies this new competitive parity or if proprietary models manage to pull ahead again.