Cheaper, Faster Chinese AI Narrows Gap With US Leaders

For businesses and developers relying on artificial intelligence, the cost and capability balance is shifting. A new release from Chinese firm Moonshot AI offers a high-performance reasoning model at a significantly lower price point than its immediate rivals. This development could pressure US-based providers to adjust their pricing structures or accelerate their own innovation cycles to maintain their competitive edge.

The model, named Kimi K3.1 Pro, has jumped to fourth place on the Artificial Analysis Intelligence Index with a score of 55. This represents an 11-point improvement over its predecessor. It currently sits just three points behind Claude Opus 5.5, which leads the proprietary models with a score of 58. If Moonshot releases the model's weights as planned by November 4, 2026, it would become the highest-scoring open-weights model available, reducing the gap between open and proprietary systems to a mere three points.

The most notable advancements are seen in agentic tasks, where the model demonstrates superior ability to perform complex, multi-step workflows. On the GDPval-AA benchmark, K3.1 Pro achieves an Elo rating of 1,846, surpassing its predecessor by 312 points. This places it nearly on par with top competitors like Claude Opus 5.5 and Muse Wonder Full. Additionally, the model shows dramatic improvements in coding, scoring 65.7 percent on Terminal-Bench 4.0, a sharp rise from the 13 percent recorded by the previous version.

Beyond raw speed, the new model offers better reliability and specialized capabilities. Moonshot reports that hallucination rates have dropped substantially, with claim-level errors falling from 23.4 percent to 8.3 percent on internal stress tests. The model also excels in 3D modeling, working directly with game engines like Unity and Unreal Engine. It scores 92.7 percent on BenchCAD when using tools, approaching the performance of leading US models. A new feature called '3D Make' allows users to preview and export 3D models directly through the platform.

Despite these performance gains, the economic structure of the model remains favorable for users. The parameter count stays at 2.8 trillion, but API prices have been reduced by 33 percent. Input costs now stand at $2 per million tokens, down from $3, while output costs drop to $10 from $15. This efficiency places the model at the top of the Pareto frontier, meaning no other model with a comparable size offers a higher intelligence score. The cost per intelligence task remains stable at $1.98, accounting for slightly longer output generation.

Contextually, this release highlights a deliberate strategy by Moonshot to balance offensive capability with safety. The company removed specific cybersecurity data from its training set for the open-weights version, limiting its ability to find vulnerabilities in software like the Linux kernel. While the uncleaned version solved 90.5 percent of such tasks, the released model solves 81.0 percent. This approach aims to allow open models to serve defensive purposes, such as auditing legacy systems, without pushing the absolute frontier of offensive cyber capabilities.

The outlook suggests a tightening race in the AI market. With the model available now through various Kimi platforms and a faster variant called K3.1 Pro Highspeed already in use, the pressure on incumbents is immediate. The combination of near-top-tier performance, lower costs, and upcoming open weights could reshape the landscape for developers choosing between proprietary and open-source solutions in the coming months.