New Chinese AI Model Narrows Gap With Top Global Rivals

Developers and businesses relying on artificial intelligence tools will see a significant drop in costs starting today. A new model from a Chinese firm is now priced 33 percent lower than its predecessor, making high-end reasoning capabilities more accessible to independent creators and smaller enterprises. This shift could reduce the financial barrier for those who prefer to manage their own computing resources rather than renting from large tech giants.

The new release, known as Kimi K3.1 Pro, has climbed to the fourth position on the Artificial Analysis Intelligence Index. It achieved a score of 55 points, marking an 11 point improvement over the previous version. This places the model just three points behind the leading proprietary systems. The only models currently scoring higher are Anthropic’s Claude Opus 5.5 at 58 points, Claude Sonnet 5.5 at 56 points, and Muse Wonder Full at 56 points.

A major development for the open source community is the planned release of the model’s weights. Moonshot AI intends to publish these files by November 4, 2026. If this schedule holds, K3.1 Pro would become the highest scoring open weights model available. This move would shrink the performance gap between open and proprietary systems from 12 points to just three, assuming no other changes occur in the meantime. This suggests a growing trend toward transparency and shared technology in the AI sector.

Performance in complex, multi-step tasks has seen the most dramatic gains. On the GDPval-AA benchmark, the new model reached an Elo rating of 1,846. This is a jump of 312 points compared to its predecessor. It now sits just behind Claude Opus 5.5 and Muse Wonder Full. In another test called Terminal-Bench 4.0, the model scored 65.7 percent. This result surpasses GPT-6 Astra, which scored 58.2 percent, and trails only slightly behind Claude Opus 5.5 at 66.4 percent. The previous version of the model had managed only 13 percent on this specific test.

Accuracy has also improved substantially. Internal testing shows a sharp decline in hallucinations, or incorrect information presented as fact. On the company’s Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent to 8.3 percent. While GPT-6 Astra still leads in this category with a 5.5 percent rate, the new model represents a significant step forward. Financial performance metrics also show strong gains, with the model topping all tested competitors on the OfficeQA Pro v2 benchmark.

The model supports a wide range of creative and technical applications. It can work with game engines such as Unity, Unreal Engine, and Godot. It is also capable of writing and running scripts for 3D software like Blender and CAD programs without needing additional connectors. On the BenchCAD benchmark, the model scored 92.7 percent when using external tools. This is close to the scores of top competitors like Claude Opus 5.5 and GPT-6 Astra. A new feature on the company’s website allows users to preview, edit, and export 3D models directly.

The underlying architecture remains consistent with the previous version. It contains 2.8 trillion total parameters, with 104 billion active during processing. The system supports a context window of one million tokens and handles both text and images. Despite the performance gains, the cost of running each benchmark task remains nearly identical to the previous model at $1.98. This stability in cost, despite higher output token generation, indicates efficient engineering.

The company has also focused on cybersecurity capabilities. Internal tests show that the model can reproduce recently disclosed vulnerabilities in software like the Linux kernel and Node.js. However, for the open weights release, the company removed specific cyber-related data from its training phases. This decision aims to balance offensive capability with defensive utility. The firm believes that open models should gradually raise the standard for security work, such as auditing legacy systems, without necessarily reaching the absolute frontier of offensive tools.