Kimi K3.1 Pro Narrows Gap With Top US Rivals, Setting New Open-Weights Benchmark

The hierarchy of artificial intelligence leadership is shifting as a Chinese contender moves aggressively into the upper echelon of global performance. Moonshot AI has unveiled Kimi K3.1 Pro, a model that has surged to fourth place on the Artificial Analysis Intelligence Index. This leap places it within striking distance of the leading proprietary systems from American competitors, signaling a tightening race among the world's most advanced reasoning engines.

The new release achieves a score of 55 points, an eleven-point increase over its predecessor. It sits just three points behind Anthropic's Claude Opus 5.5, which holds the top spot with 58 points. Only Claude Sonnet 5.5 and Muse Wonder Full, both scoring 56, currently rank higher. If Moonshot adheres to its plan to release the model weights by November 4, 2026, K3.1 Pro will become the highest-scoring open-weights model available. This move would drastically shrink the performance divide between open and proprietary models from twelve points to just three, assuming no other major releases intervene.

Performance gains are most evident in agentic tasks and coding capabilities. On the GDPval-AA benchmark, the model reaches 1,846 Elo, a substantial rise from its predecessor. It trails Claude Opus 5.5 by only twenty points on this specific metric. In terminal-based evaluations, K3.1 Pro reports a success rate of 65.7 percent, slightly behind the 66.4 percent achieved by Claude Opus 5.5. Hallucination rates have also dropped significantly, with claim-level errors falling from 23.4 percent to 8.3 percent on internal stress tests, though GPT-6 Astra still maintains an edge with a 5.5 percent error rate.

The model retains the same architectural size as its predecessor, featuring 2.8 trillion total parameters and 104 billion active ones. Despite this identical scale, Moonshot has reduced API prices by 33 percent. Input costs now stand at $2 per million tokens, while output costs are $10. A new high-speed variant is also available, operating two to three times faster than the standard version. Both models are accessible through various platforms including Kimi.ai and the Kimi API.

A notable focus in this release is the handling of cybersecurity capabilities. Moonshot removed cyber-related data from the training corpus for the open-weights version to balance offensive and defensive potential. The uncleaned model solved 90.5 percent of tasks on an internal benchmark, while the cleaned version solved 81.0 percent. The company argues that open models should not dominate the absolute frontier of offensive capabilities but should still be strong enough for defensive auditing.

Questions remain regarding how the broader market will react to these price cuts and the impending open-weights release. The close proximity to top proprietary models challenges the notion that closed systems hold an insurmountable advantage. As the release date for the weights approaches, the industry will watch closely to see if this new standard holds up against emerging competitors.