AI Rivals: Moonshot's New Model Narrows Gap With Top Proprietary Systems

A new Chinese AI model is rapidly closing the distance to the industry's leading proprietary systems.

Moonshot AI has launched Kimi K3.1 Pro, a post-trained update to its previous version. The release features significantly more compute during the training phase. This effort resulted in an 11-point jump on the Artificial Analysis Intelligence Index. The model now scores 55 points, placing it in fourth position. Only three models rank higher: Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. Moonshot plans to release the model weights by November 4, 2026. If this happens, it will become the highest-scoring open-weights model available. That move would shrink the performance gap between open and proprietary models from 12 points to just three.

The most notable improvements appear in agentic tasks. On the GDPval-AA benchmark, the model achieves 1,846 Elo. This is 312 points above its predecessor. It now trails only Claude Opus 5.5 and Muse Wonder Full in that specific metric. Coding capabilities have also seen a massive leap. The model scores 65.7 percent on Terminal-Bench 4.0. That figure exceeds GPT-6 Astra and sits barely behind Claude Opus 5.5. The previous version of Kimi K3 managed only 13 percent on the same test. Hallucination rates have dropped sharply as well. Internal tests show claim-level errors falling from 23.4 percent to 8.3 percent. While GPT-6 Astra still holds an edge at 5.5 percent, Moonshot notes this is a stress test rather than a general error measure.

The model also excels in 3D modeling and specific professional tasks. It can interact directly with game engines like Unity and Unreal Engine. On the BenchCAD benchmark, it scores 92.7 percent when using tools. That is close to the performance of top competitors. Without tools, the score is 72.8 percent, nearly matching the leader. Moonshot states the model is particularly strong at parametric modeling for mechanical parts and architecture. It can write and run scripts for software like Blender without needing extra connectors. A new feature called "3D Make" allows users to preview and export these models directly on the platform.

Despite the performance gains, the model keeps the same size as its predecessor. It has 2.8 trillion total parameters, with 104 billion active. The context window spans one million tokens and handles both text and images. API prices have dropped by 33 percent. Input costs are now $2 per million tokens, down from $3. Output costs are $10, reduced from $15. Cache hits cost $0.20, down from $0.30. This makes the model highly competitive on the Pareto frontier. No other model with a similar parameter count scores higher on the intelligence index.

Moonshot also addressed cybersecurity capabilities in its technical documentation. The company observed that vulnerability research skills grew as post-training scaled up. To prepare for the open-weights release, they removed cyber-related data from both training phases. This adjustment reduced the model's ability to solve specific security tasks from 90.5 percent to 81.0 percent. While this is lower than some competitors, it remains far above the previous version. The company believes open models should not push the absolute frontier of offensive cyber capabilities. However, they want to gradually improve these skills to help with defensive tasks like auditing legacy systems.