Kimi K3.1 Pro Leads Agentic Coding but Trails on AA-3D 3D Scene Benchmark

Kimi K3.1 Pro ranks among the leading models in agentic knowledge work and coding on GDPval-AA and Terminal-Bench 4.0, while early signs indicate it will rank lower on AA-3D, the benchmark for building interactive 3D scenes. Preference votes for its AA-3D Elo are still being collected, and its predecessor Kimi K3 placed 12th of 22 models on that benchmark. One AA-3D scene shows a 492 m mountain-bike descent with berms, roots, a rock garden, a drop and a jump, with a demo run and user control of steering, pedaling and braking.
Event version 1.
Key insights
Strong agentic and coding performance
Kimi K3.1 Pro is among the leading models in agentic knowledge work and coding. It ranks among the top models on GDPval-AA and Terminal-Bench 4.0.
Lower expected AA-3D ranking
Early signs suggest Kimi K3.1 Pro will rank lower on AA-3D, the benchmark for how well AI models build interactive 3D scenes. Preference votes for its AA-3D Elo are still being collected.
Predecessor placed 12th of 22
Kimi K3, the predecessor model, is 12th of 22 models on AA-3D.
Interactive mountain-bike scene demo
One AA-3D scene is a 492 m mountain-bike descent with berms, roots, a rock garden, a drop and a jump. A demo run rides it from top to bottom, and users can take over steering, pedaling and braking at any point.