Moonshot AI's Kimi K3.1 Pro Narrows Gap to Anthropic's Claude Opus 5.5

Chinese AI company Moonshot AI has released a new reasoning model, Kimi K3.1 Pro, which has immediately secured the fourth position on the Artificial Analysis Intelligence Index with a score of 55. The model's performance places it just 3 points behind the top-scoring proprietary models currently available.

Three points separate K3.1 Pro from Claude Opus 5.5, which leads the index at 58. Other models scoring higher than K3.1 Pro include Claude Sonnet 5.5 and Muse Wonder Full, both at 56. This latest release from Moonshot AI represents a significant jump in capability from its predecessor, the Kimi K3, which had a score 11 points lower.

The model's most substantial improvements are in agentic tasks, coding, and 3D modeling. On the Terminal-Bench 4.0 benchmark, K3.1 Pro scores 65.7 percent, a notable increase from the 13 percent its predecessor achieved. Hallucination rates have also been reduced, falling from 23.4 percent with K3 to 8.3 percent on an internal benchmark, a figure the company characterizes as a stress test rather than a measure of general error rates.

K3.1 Pro also demonstrates significant capabilities in 3D modeling and coding. According to Moonshot, it can work directly with game engines like Unity and Unreal Engine. On the BenchCAD benchmark, it scores 92.7 percent with tools, a result close to Claude Opus 5.5's 96.2 percent. Without tools, K3.1 Pro scores 72.8 percent, nearly matching Claude Opus 5.5's score of 73.0 percent.

The model's architecture remains consistent with its predecessor, maintaining 2.8 trillion total parameters with 104 billion active ones. However, API prices have been reduced by 33 percent. Input costs are now $2 per million tokens, down from $3, while output costs are $10 per million tokens, down from $15.

Moonshot has also addressed cybersecurity concerns in this release. The company removed cyber-related data from the training process for the open-weights version of the model. Following this data cleaning, K3.1 Pro scores 81.0 percent on an internal benchmark that reproduces recently disclosed vulnerabilities, a decrease from the 90.5 percent score of an uncleaned version.

Plans are in place for Moonshot to release the model weights by November 4, 2026. If this release proceeds as scheduled, K3.1 Pro would become the highest-scoring open-weights model, which would reduce the gap between open and proprietary models from 12 points to 3. The model is available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

Kimi K3.1 Pro is a post-trained version of its predecessor K3 with significantly more post-training compute. Alongside K3.1 Pro, Moonshot is shipping a variant called K3.1 Pro Highspeed that runs two to three times faster, according to the company.