Moonshot AI releases Kimi K3.1 Pro to close gap with top rivals
The new reasoning model jumps to fourth place on major intelligence benchmarks while drastically cutting hallucination rates.
By Aiko Soto ·
Moonshot AI has unveiled Kimi K3.1 Pro, a new reasoning model that secures the fourth spot on the Artificial Analysis Intelligence Index with a score of 55. This release marks a significant leap for the Chinese firm, as the new model improves upon its predecessor by 11 points. It now sits just three points behind the leading proprietary systems. Only Anthropic's Claude Opus 5.5 and Claude Sonnet 5.5, along with Muse Wonder Full, currently outscore it. If the company follows through on its plan to release model weights by November 4, 2026, K3.1 Pro will become the highest-scoring open-weights model available. This move would narrow the performance gap between open and proprietary models from 12 points to just three.
The jump in rankings follows a period of rapid iteration. The new system is a post-trained version of the earlier Kimi K3 that utilized substantially more compute. Moonshot is simultaneously offering a Highspeed variant that the company says operates two to three times faster. Users can access both versions through Kimi.ai, Kimi Work, Kimi Code, and the company's API platform.
Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, the new model achieved an Elo rating of 1,846. This figure is 312 points higher than the previous version and trails only Claude Opus 5.5 and Muse Wonder Full. On Terminal-Bench 4.0, Moonshot reports a success rate of 65.7 percent. This result beats GPT-6 Astra, which scored 58.2 percent, and stays within a narrow margin of Claude Opus 5.5 at 66.4 percent. The earlier K3 model managed only 13 percent on that same test.
Reliability has also improved markedly. Internal tests show a sharp drop in hallucinations. Claim-level errors fell from 23.4 percent with the old model to 8.3 percent with the new one. While GPT-6 Astra still leads this metric at 5.5 percent, Moonshot notes the benchmark serves as a stress test. In finance, the model gained 7.2 points in modeling and 11.6 points in research. Moonshot claims it now tops all tested models on OfficeQA Pro v2.
The system also demonstrates stronger capabilities in 3D modeling and coding. It integrates with Unity, Unreal Engine, and Godot, and can script for Blender and CAD software without external connectors. On the BenchCAD benchmark, it scores 92.7 percent with tools, slightly below its top competitors. A new 3D Make feature on Kimi.com allows users to preview and export models.
The architecture remains unchanged from the previous version, featuring 2.8 trillion total parameters with 104 billion active. It supports a one-million-token context window. API prices have dropped by 33 percent, with input costs at $2 per million tokens. Despite generating 14 percent more output, the cost per intelligence index task remains nearly identical to the previous model.
Security experts will watch the open-weights release closely. Moonshot states that vulnerability research capabilities grew during training but were scrubbed for the public release. On an internal benchmark, the cleaned model solved 81.0 percent of tasks, compared to 90.5 percent for the uncleaned version. The company aims to gradually raise the ceiling for open models while supporting defensive security work.
