AI Race: Moonshot's Kimi K3.1 Pro Narrows Gap With Top U.S. Models

The distance between the leading proprietary artificial intelligence models and their open-weights competitors has never been smaller, but a new release from Moonshot AI is poised to compress that margin further. By landing at number four on the Artificial Analysis Intelligence Index, the company’s latest reasoning model has effectively closed the gap with the industry’s top performers, challenging the dominance of American rivals in a highly competitive landscape.

Kimi K3.1 Pro, a post-trained evolution of the previous K3 iteration, achieved a score of 55 on the Intelligence Index. This represents an 11-point leap from its predecessor and places it directly behind Claude Opus 5.5, which holds the top spot with 58 points, as well as Claude Sonnet 5.5 and Muse Wonder Full, both of which scored 56. The ranking underscores how rapidly the benchmark landscape is shifting, with only three points separating the Chinese model from the proprietary leader.

The most significant improvements for K3.1 Pro are visible in agentic tasks, where the model demonstrates substantial gains in autonomy and execution. On the GDPval-AA benchmark, the model reached an Elo rating of 1,846, a jump of 312 points from the previous version. It now trails only Claude Opus 5.5 and Muse Wonder Full in this specific metric. Furthermore, on Terminal-Bench 4.0, Moonshot reported a performance rate of 65.7 percent, surpassing GPT-6 Astra and falling only slightly behind the top-ranked Claude model.

Beyond raw task performance, the new model shows marked improvements in reliability. According to Moonshot, the rate of claim-level hallucinations dropped from 23.4 percent to 8.3 percent on the company’s internal Kimi Hallu Bench Hard. While GPT-6 Astra maintains a slight edge with a 5.5 percent rate, Moonshot notes that this benchmark serves as a stress test rather than a measure of general error rates. The model also posted gains in financial modeling and research, and it currently leads all tested models on the OfficeQA Pro v2 benchmark for general office tasks.

In specialized technical fields, K3.1 Pro integrates directly with major game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, the model achieved a 92.7 percent score when using tools, approaching the performance of both Claude Opus 5.5 and GPT-6 Astra. Moonshot highlights the model’s strength in parametric modeling for mechanical parts and architecture, noting that it can write and execute scripts for Blender and CAD software without the need for external MCP connectors. A new feature called 3D Make on Kimi.com allows users to preview, edit, and export these models.

Despite the performance jump, the underlying architecture remains consistent with the previous version. K3.1 Pro retains a total parameter count of 2.8 trillion, with 104 billion active parameters, and supports a context window of one million tokens for both text and image inputs. Artificial Analysis places the model at the top of the Pareto frontier, indicating that no other model with a comparable parameter count scores higher on the Intelligence Index. This efficiency allows Moonshot to reduce API prices by 33 percent, with input costs dropping to $2 per million tokens and output costs falling to $10.

Moonshot has also addressed cybersecurity considerations in its technical documentation. The company observed that vulnerability research capabilities increased as post-training scaled, leading to a strategic decision to strip cyber-related data from the training corpus for the open-weights release. On the internal xdev-bench-hard test, which replicates recent vulnerabilities in software like the Linux kernel, the uncleaned model solved 90.5 percent of tasks, while the cleaned K3.1 Pro solved 81.0 percent. This is slightly lower than DeepSeek-V4.1-Flash, which reached 85.7 percent, but significantly higher than the previous Kimi K3, which scored 33.3 percent.

The company plans to release the model weights by November 4, 2026. If this schedule holds, K3.1 Pro will become the highest-scoring open-weights model available. This move would shrink the performance gap between open and proprietary models from 12 points to just three, assuming no other major releases occur before then. A variant called K3.1 Pro Highspeed, which runs two to three times faster, is also available now across Kimi’s various platforms and API services.