Moonshot AI's Kimi K3.1 Pro Narrows Gap With Leading US Rivals
A new Chinese reasoning model has surged in performance rankings, matching top proprietary systems on key benchmarks while reducing API costs by a third.
By Carmen Wojcik ·
The race to build the most capable artificial intelligence models has intensified as Chinese firms close the distance to American leaders. On Thursday, Moonshot AI introduced Kimi K3.1 Pro, a post-trained version of its previous model that has quickly climbed to fourth place on the Artificial Analysis Intelligence Index.
The new model scores 55 points, an 11-point improvement over its predecessor, Kimi K3. This leap places it just three points behind the leading proprietary models, including Anthropic’s Claude Opus 5.5 and Claude Sonnet 5.5. If the company follows its plan to release the model weights by November 4, 2026, K3.1 Pro would become the highest-scoring open-weights model available. Such a move would shrink the performance gap between open and proprietary systems from 12 points to just three.
Significant gains are visible in agentic tasks and coding. On the GDPval-AA benchmark, the model achieves an Elo rating of 1,846, which is 312 points higher than the previous version. It trails only Claude Opus 5.5 and Muse Wonder Full in this category. In coding tests, specifically Terminal-Bench 4.0, K3.1 Pro reports a success rate of 65.7 percent, surpassing GPT-6 Astra and coming within 0.7 percent of Claude Opus 5.5. The model also demonstrates a sharp reduction in hallucinations, with claim-level errors dropping from 23.4 percent to 8.3 percent on internal stress tests.
Beyond general reasoning, the model shows strong capabilities in 3D modeling and finance. It can interact directly with game engines like Unity and Unreal Engine, scoring 92.7 percent on the BenchCAD benchmark when using tools. Moonshot also launched a feature on its platform that allows users to preview and export 3D models. In financial modeling and research tasks, the model recorded double-digit point improvements over its predecessor, topping all tested models on office task benchmarks.
Despite the performance boost, the model maintains the same architecture size, with 2.8 trillion total parameters and 104 billion active ones. However, the cost of using the model has decreased. Input prices have dropped to $2 per million tokens, and output prices to $10, representing a 33 percent reduction from previous rates. This makes the model efficient for developers, with the cost per intelligence task remaining nearly identical to the previous version despite the model generating 14 percent more output tokens.
Moonshot has also addressed cybersecurity concerns by modifying the training data for the open-weights release. The company removed cyber-related data from both training phases to prevent the model from pushing the absolute frontier of offensive capabilities. On internal benchmarks that reproduce recent software vulnerabilities, the cleaned model solves 81.0 percent of tasks, down from 90.5 percent for the uncleaned version. This approach aims to balance offensive potential with defensive utility, such as auditing legacy systems.
The release marks a significant step for open-source AI, as the new model is available now through various Moonshot platforms. A faster variant, K3.1 Pro Highspeed, is also offered, running two to three times quicker than the standard version. As the weights are set for release next month, the industry will watch closely to see if this model reshapes the competitive landscape between open and proprietary AI systems.
