Kimi K3.1 Pro Scores 55, Just Three Points Behind Top AI Rivals

Moonshot AI released Kimi K3.1 Pro on Thursday, a reasoning model that scored 55 on the Artificial Analysis Intelligence Index. That result places the system in fourth position, directly behind Anthropic’s Claude Opus 5.5, which holds 58 points, and Claude Sonnet 5.5 and Muse Wonder Full, both at 56. The new version also runs 33% cheaper for developers.

The jump represents an 11-point increase over the predecessor, Kimi K3. According to Artificial Analysis, the model sits on the Pareto frontier, meaning no other model of its 2.8 trillion parameter size scores higher. The company used substantially more post-training compute for K3.1 Pro while keeping the architecture identical to K3, including the 104 billion active parameters and one-million-token context window.

Performance gains were most visible in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached 1,846 Elo, a 312-point leap from K3. It trails only Claude Opus 5.5 and Muse Wonder Full. In terminal coding, Moonshot reports a 65.7% score on Terminal-Bench 4.0, surpassing GPT-6 Astra at 58.2% and landing just below Claude Opus 5.5 at 66.4%. K3 previously managed only 13% on that test.

Hallucination rates dropped sharply as well. On Moonshot’s internal Kimi Hallu Bench Hard, claim-level errors fell from 23.4% for K3 to 8.3% for the new model. The company notes that GPT-6 Astra still leads this specific metric at 5.5%. Financial benchmarks also improved, with gains of 7.2 points in modeling and 11.6 points in research.

The model supports 3D modeling and coding for engines like Unity and Unreal. It achieves a 92.7% score on BenchCAD with tools, slightly behind Claude Opus 5.5’s 96.2%. Moonshot is also adding a “3D Make” feature to Kimi.com for previewing and exporting models. Both the standard and a faster “Highspeed” variant are available via Kimi.ai and the Kimi API.

API pricing has dropped significantly. Input tokens now cost $2 per million, output tokens $10, and cache hits $0.20. Although the model generates 14% more output tokens, the cost per Intelligence Index task is $1.98, nearly identical to K3’s $2.00. Moonshot plans to publish open weights by November 4, which would make it the highest-scoring open-weights model available.

For the open release, Moonshot removed cyber-related data from training to manage offensive capabilities. On its internal xdev-bench-hard test, the uncleaned model solved 90.5% of vulnerability tasks, dropping to 81.0% after cleaning. This compares to 85.7% for DeepSeek-V4.1-Flash and 33.3% for Kimi K3.