New Chinese AI Model Ranks Fourth Globally

An 11-point jump in the latest intelligence benchmarks has propelled a new Chinese artificial intelligence model into the global top four. Kimi K3.1 Pro, developed by Moonshot AI, now scores 55 points on the Artificial Analysis Intelligence Index. This performance places it just three points behind the leading proprietary models, marking a significant shift in the competitive landscape for large language models.

The release builds on the previous Kimi K3 architecture, which scored 44 points on the same index. By applying significantly more post-training compute, Moonshot AI has transformed the model's capabilities. As of this week, only three models rank higher: Anthropic’s Claude Opus 5.5 with 58 points, Claude Sonnet 5.5 with 56, and Muse Wonder Full with 56. If the company releases the model weights by the planned date of November 4, 2026, Kimi K3.1 Pro will become the highest-scoring open-weights model available.

The primary improvements are visible in agentic tasks, where the model demonstrates substantial gains over its predecessor. On the GDPval-AA benchmark, Kimi K3.1 Pro achieves an Elo score of 1,846, a 312-point increase from the previous version. It also records a 65.7 percent success rate on Terminal-Bench 4.0, a metric where the older K3 model managed only 13 percent. These figures place it close to the industry leader, Claude Opus 5.5, which scores 66.4 percent on the same test.

Reliability has also improved, with internal tests showing a sharp drop in hallucination rates. According to Moonshot AI, claim-level hallucinations fell from 23.4 percent in the previous version to 8.3 percent in K3.1 Pro. While the company notes this is a stress test rather than a general error measure, it still trails GPT-6 Astra, which records a 5.5 percent rate on the same benchmark. The model also shows stronger performance in financial modeling and research, as well as topping general office task evaluations.

Specialized capabilities in 3D modeling and coding have expanded considerably. The model can now work directly with game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent when using tools, a result close to top competitors. Without tools, it achieves 72.8 percent, nearly matching Claude Opus 5.5. Moonshot AI states the model is particularly effective at parametric modeling of mechanical parts and architectural designs, and it can write scripts for software like Blender without additional connectors.

Despite the performance boost, the model maintains the same structural dimensions as its predecessor, with 2.8 trillion total parameters and 104 billion active ones. The context window supports one million tokens and handles both text and images. For developers, the cost of using the model has decreased significantly. Input prices have dropped to $2 per million tokens from $3, while output costs are now $10 per million tokens, down from $15. This represents a 33 percent reduction in API pricing.

Moonshot AI has also addressed cybersecurity concerns by removing specific cyber-related data from the training phases for the open-weights release. This adjustment reduced the model's ability to solve recent vulnerability tasks from 90.5 percent to 81.0 percent on internal tests. The company aims to balance offensive capabilities with defensive utility, noting that open-weights models should not push the absolute frontier of offensive cyber capabilities while still serving purposes like auditing legacy systems.

The new model is available immediately through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. A variant called K3.1 Pro Highspeed is also shipping, which runs two to three times faster than the standard version. With the weights release scheduled for early November, the gap between open and proprietary models is expected to shrink from 12 points to just three, assuming no other major releases occur in the interim.

This development underscores the rapid pace of iteration in the AI sector, where Chinese firms are increasingly matching or exceeding Western benchmarks in specific technical domains while offering lower costs. The strategic decision to release weights soon may further influence how developers choose between closed and open architectures for their applications.