Moonshot AI Model Climbs to 55 on Intelligence Index, Trailing Claude Opus 5.5 by Three Points
The new Kimi K3.1 Pro delivers significant improvements in agentic capabilities and reduced pricing ahead of its scheduled open-weights release.
By Madison Lewis ·
Moonshot AI has scored 55 points on the Artificial Analysis Intelligence Index with its new reasoning model, Kimi K3.1 Pro, positioning the system as the fourth-ranked model available today.
The company stated that the release is a post-trained iteration of its predecessor, Kimi K3, built with significantly more compute during the post-training phase. According to Artificial Analysis, this update resulted in an 11-point increase in the index score over the prior version. The model currently ranks behind Claude Opus 5.5 at 58 points, Claude Sonnet 5.5 at 56, and Muse Wonder Full at 56.
A scheduled release of model weights by November 4, 2026, would make K3.1 Pro the highest-scoring open-weights model on the market. Provided no other system surpasses that benchmark before the date, the release would reduce the performance gap between open-weights and proprietary models from 12 points to 3.
The model's performance gains are concentrated in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached an Elo of 1,846, compared to 1,534 for K3, placing it just below Claude Opus 5.5 at 1,866. On Terminal-Bench 4.0, Moonshot reported a score of 65.7 percent, an increase from K3's 13 percent. Moonshot also reported that hallucination rates on its internal Kimi Hallu Bench Hard test fell from 23.4 percent to 8.3 percent.
The company's internal benchmarks for financial modeling and financial research increased by 7.2 points and 11.6 points, respectively. Regarding general office tasks, the model is reported to lead all tested systems on OfficeQA Pro v2.
K3.1 Pro retains the architecture specifications of K3, featuring 2.8 trillion total parameters and 104 billion active parameters, with a one-million-token context window. Artificial Analysis places the model at the top of the Pareto frontier, indicating that no model with a comparable parameter count achieves a higher score on the Intelligence Index.
The cost to access the model through the API has decreased alongside these performance updates. Input prices are set at $2 per million tokens, down from $3, while output prices are set at $10, down from $15. Moonshot reports that the total cost per Intelligence Index task remains effectively unchanged at $1.98, compared to $2.00 for K3, as the new model generates 14 percent more output tokens per task.
For the upcoming open-weights release, Moonshot has stripped cyber-related data from both the mid-training and reinforcement learning phases. On the internal xdev-bench-hard test, which reproduces recently disclosed vulnerabilities in systems such as the Linux kernel, Node.js, and JavaScriptCore, the cleaned version of K3.1 Pro solves 81.0 percent of tasks, compared to 90.5 percent for the uncleaned version. DeepSeek-V4.1-Flash reaches 85.7 percent on the same test.
