Moonshot AI Narrows Gap With Top Global Models

The race for artificial intelligence supremacy has just accelerated in the Chinese tech sector.

Moonshot AI has officially released Kimi K3.1 Pro, a heavily optimized successor to its previous model that immediately carved out a significant position in the global market. According to Artificial Analysis, the new model scored 55 points on the Intelligence Index, placing it in fourth position globally. This represents an 11-point improvement over its predecessor, Kimi K3, and puts it within striking distance of the leading proprietary systems. Only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full currently hold higher rankings. If Moonshot adheres to its plan to release the model weights by early November, K3.1 Pro would become the highest-performing open-weights model available, effectively reducing the performance gap between open and closed systems from twelve points to just three.

The most dramatic improvements in the new release are evident in agentic tasks, where the model demonstrates a substantial leap in capability. On the GDPval-AA benchmark, Kimi K3.1 Pro achieved an Elo score of 1,846, a rise of 312 points from the previous version. This performance places it just behind the top contenders, such as Claude Opus 5.5 and Muse Wonder Full. In coding-specific tests, the model also showed remarkable progress, scoring 65.7 percent on Terminal-Bench 4.0. This figure significantly exceeds the 13 percent recorded by the original K3 and surpasses GPT-6 Astra, trailing only slightly behind Claude Opus 5.5. The company notes that these gains are accompanied by a sharp reduction in hallucination rates, with claim-level errors dropping from 23.4 percent to 8.3 percent on internal stress tests, though GPT-6 Astra still maintains a lower error rate at 5.5 percent.

Beyond general reasoning, the model exhibits strong capabilities in specialized fields like finance and three-dimensional modeling. In internal financial benchmarks, K3.1 Pro posted gains in both modeling and research categories, while topping all tested models on OfficeQA Pro v2 for general office tasks. For developers and designers, the model can interact directly with game engines such as Unity, Unreal, and Godot. On the BenchCAD benchmark, it scored 92.7 percent when using tools, a figure that closely mirrors the performance of top-tier proprietary models. Moonshot has also introduced a new feature on its platform that allows users to preview, edit, and export 3D models, highlighting the system's strength in parametric modeling for mechanical and architectural applications.

Despite the performance boost, Moonshot has kept the architectural footprint identical to the previous version, maintaining 2.8 trillion total parameters with 104 billion active ones. The context window remains at one million tokens, supporting both text and image inputs. However, the economic structure of the model has shifted in favor of developers. API prices have dropped by 33 percent, with input costs falling to two dollars per million tokens and output costs to ten dollars. Although the model generates 14 percent more output tokens, the cost per intelligence task remains nearly identical to the previous version at $1.98. This positions the model at the top of the Pareto frontier, meaning no other model of similar size scores higher on the Intelligence Index.

A notable aspect of the release is the company’s approach to cybersecurity capabilities. Moonshot observed that vulnerability research skills grew as post-training scaled up, prompting a deliberate decision to strip cyber-related data from the training corpus for the open-weights release. On an internal benchmark that reproduces known vulnerabilities in software like the Linux kernel, the uncleaned version solved 90.5 percent of tasks, while the cleaned K3.1 Pro solved 81.0 percent. While this is higher than the previous Kimi K3 score of 33.3 percent, it remains below the 85.7 percent achieved by DeepSeek-V4.1-Flash. The company argues that open models should not push the absolute frontier of offensive cyber capabilities but should still be capable enough to assist in defensive auditing, a stance that contrasts with recent findings by competitors regarding the exploit-writing abilities of other open models.

The release also includes a high-speed variant of the model, which Moonshot claims runs two to three times faster than the standard version. Both versions are now available through various platforms, including the company’s API and work tools. As the open-weights release approaches, the market is watching closely to see if the three-point gap to the proprietary leaders holds or if new entrants emerge to challenge the current hierarchy.