Moonshot AI's Kimi K3.1 Pro Narrows Gap With Top US Models

Moonshot AI has released Kimi K3.1 Pro, a reasoning model that has climbed to fourth place on the Artificial Analysis Intelligence Index with a score of 55. This significant leap places it within striking distance of the current leaders, specifically Anthropic’s Claude Opus 5.5, which holds the top spot with a score of 58. The new model marks a substantial improvement over its predecessor, Kimi K3, representing an 11-point increase in overall intelligence metrics. If Moonshot releases the model’s weights as scheduled by November 4, 2026, it would become the highest-performing open-weights model available, shrinking the performance gap between open and proprietary systems from 12 points to just three.

The upgrade focuses heavily on agentic capabilities, coding, and 3D modeling. On the GDPval-AA benchmark, K3.1 Pro achieves an Elo rating of 1,846, a 312-point jump from its predecessor. In Terminal-Bench 4.0 testing, the model records a 65.7 percent success rate, surpassing GPT-6 Astra and narrowly trailing Claude Opus 5.5. Hallucination rates have also decreased significantly; on Moonshot’s internal stress test, claim-level hallucinations dropped from 23.4 percent to 8.3 percent, though GPT-6 Astra still maintains a lower rate of 5.5 percent.

Beyond text, the model demonstrates strong performance in technical and creative domains. It can interact directly with game engines such as Unity, Unreal Engine, and Godot to manage both minor repairs and full project builds. In 3D modeling, it scores 92.7 percent on the BenchCAD benchmark when using tools, closely following Claude Opus 5.5. Moonshot notes that the model is particularly effective at parametric modeling for mechanical parts and architecture, and it can write scripts for Blender and CAD software without requiring specific connectors.

Economically, the new release is more accessible than its predecessor. While the parameter count remains at 2.8 trillion, with 104 billion active parameters, the API pricing has been reduced by 33 percent. Input costs have fallen to $2 per million tokens, and output costs are now $10 per million. Cache hits are priced at $0.20 per million tokens. According to Artificial Analysis, the model sits on the Pareto frontier, meaning no other model with a similar parameter count achieves a higher score on the intelligence index.

Moonshot also addressed cybersecurity considerations in its technical documentation. The company observed that vulnerability research capabilities grew as post-training scaled, prompting them to strip cyber-related data from both training phases for the open-weights release. On an internal benchmark that reproduces recent vulnerabilities, the uncleaned version solved 90.5 percent of tasks, whereas the cleaned version solves 81.0 percent. This approach aims to balance offensive capability with defensive utility, as open models are increasingly used for auditing legacy systems.

The model is available immediately through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. A variant called K3.1 Pro Highspeed, which operates two to three times faster, is also part of the rollout. The release includes a context window of one million tokens and supports both text and image inputs. As the industry continues to track the rapid evolution of Chinese AI models, K3.1 Pro represents a notable step toward parity with leading proprietary systems from the United States.

The competitive landscape in artificial intelligence is shifting as open-weights models close the performance gap with their proprietary counterparts. Moonshot’s decision to lower prices while improving core capabilities suggests a strategy to capture a larger share of the developer market. The upcoming release of the model weights in November will be a key indicator of how quickly open-source communities can adopt and build upon this new standard.