Moonshot AI Unveils Kimi K3.1 Pro, Closing Gap With Top AI Models
The new model from Moonshot AI has achieved a high ranking on a major intelligence index, coming within a few points of leading proprietary systems.
By Tunde Adeyemi ·
Chinese AI company Moonshot AI has released a new reasoning model, Kimi K3.1 Pro, which immediately ranks as one of the world's most capable AI systems.
Kimi K3.1 Pro has scored 55 points on the Artificial Analysis Intelligence Index, placing it fourth overall and just three points behind the top proprietary models, including Claude Opus 5.5. This marks an 11-point increase over its predecessor, Kimi K3. The model is a post-trained version of Kimi K3, developed with significantly more post-training compute.
Moonshot AI plans to release the model's weights by November 4, 2026. If this schedule holds, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This would reduce the performance gap between open and proprietary models from 12 points to just 3 points, provided no other model changes its score before then. Alongside K3.1 Pro, a variant named K3.1 Pro Highspeed is also being offered, which the company says is two to three times faster. Both models are accessible now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
The model shows major gains in its ability to handle agentic tasks. On the GDPval-AA benchmark, it reaches an Elo score of 1,846, a 312-point increase over K3, placing it just behind Claude Opus 5.5. On Terminal-Bench 4.0, K3.1 Pro scores 65.7 percent, surpassing GPT-6 Astra's 58.2 percent and approaching Claude Opus 5.5's 66.4 percent. The previous K3 model scored only 13 percent on that same test. Hallucination rates have also decreased significantly, dropping from 23.4 percent to 8.3 percent on an internal stress test.
Improvements are also evident in coding and 3D modeling. The model can interact with game engines like Unity, Unreal Engine, and Godot to perform tasks ranging from simple fixes to building complete projects. On the BenchCAD benchmark for 3D modeling, it scores 92.7 percent when using tools, a result close to Claude Opus 5.5 and GPT-6 Astra. The model's size remains the same as its predecessor, with 2.8 trillion total parameters and 104 billion active parameters, and it supports a context window of one million tokens.
Despite the performance gains, the cost to use the model has gone down. API prices have been reduced by 33 percent, with input costs dropping to $2 per million tokens and output costs to $10 per million tokens. For its open-weights release, Moonshot AI removed cyber-related data from its training process to limit offensive capabilities, though the model retains strong skills for defensive security tasks.
Moonshot AI will continue to provide updates on the model's release schedule.
Source: The Decoder
