Kimi K3.1 Pro challenges top AI models

A new Chinese AI model, Kimi K3.1 Pro, has risen to fourth place on the Artificial Analysis Intelligence Index, nearly matching the top proprietary models. The model shows significant improvements in agentic tasks, coding, and 3D modeling. These enhancements, combined with a 33 percent drop in API prices, position it as a strong competitor in the AI landscape.

Kimi K3.1 Pro's performance on the Artificial Analysis Intelligence Index reflects its major gains. It scores 55 points, an 11-point increase over its predecessor, placing it just behind Claude Opus 5.5. If Moonshot AI releases the model weights as planned by November 2026, it would become the highest-scoring open-weights model available, significantly narrowing the gap between open and proprietary models.

The model's strongest improvements are seen in agentic tasks, where it outperforms many leading models. It reaches 1,846 Elo on the GDPval-AA benchmark, 312 points above its predecessor. On Terminal-Bench 4.0, it scores 65.7 percent, just behind Claude Opus 5.5. Its hallucination rates also dropped sharply, showing it to be more accurate than before.

Kimi K3.1 Pro excels in 3D modeling and coding, working directly with game engines like Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent with tools, close to Claude Opus 5.5. The model's capabilities in parametric modeling of mechanical parts, architectural models, and configurable products, along with its ability to write and run scripts for Blender and CAD software, make it a powerful tool for developers and designers.

Despite its performance, the model maintains the same dimensions as its predecessor, with 2.8 trillion total parameters and 104 billion active ones. Token prices have dropped by 33 percent, making it more cost-effective. Cybersecurity is a key focus, with Moonshot stripping cyber-related data from the training phases to prevent the model from pushing the absolute frontier of offensive cyber capabilities.

Moonshot is committed to gradually raising what open models can do, as they serve defensive purposes like auditing legacy systems. The company believes that even models with lower scores on the AA Index can write exploits at near-Claude Mythos levels, showing that the competition in AI is both fierce and evolving.