New Chinese AI model narrows the gap with top global rivals
Moonshot AI's latest release scores nearly as high as leading proprietary models while offering significant cost reductions and improved coding capabilities.
By Mariana Delgado ·
The race to build the most capable artificial intelligence models is intensifying, with new contenders challenging long-standing leaders from the United States. On October 23, 2026, Moonshot AI announced the release of Kimi K3.1 Pro, a new reasoning model that has rapidly climbed the global rankings to sit just behind the current market leaders.
This development marks a significant shift in the competitive landscape. Previously, the top-tier proprietary models maintained a comfortable lead over open-weights alternatives. However, the latest benchmark results suggest that this hierarchy is beginning to blur. The new model achieves a score of 55 on the Artificial Analysis Intelligence Index, an 11-point improvement over its predecessor. This places it in fourth position globally, trailing only three specific models: Claude Opus 5.5, which leads with 58 points, and Claude Sonnet 5.5 and Muse Wonder Full, both at 56 points.
The trajectory of this model reflects a broader trend in the industry. Moonshot AI developed K3.1 Pro by applying significantly more post-training compute to the existing Kimi K3 architecture. This additional training focused heavily on agentic tasks, coding, and 3D modeling. The result is a system that performs notably better in real-world applications, particularly in autonomous software execution. According to the company, the model’s ability to handle complex, multi-step tasks has improved dramatically compared to the previous version.
Looking ahead, the most significant change may come in November 2026. Moonshot plans to release the full model weights by November 4. If this timeline holds, K3.1 Pro would 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 in the interim. Such a reduction would make high-performance AI more accessible to developers who prefer open-source tools.
Beyond raw intelligence scores, the new model offers substantial economic incentives for users. While the parameter count remains unchanged at 2.8 trillion total parameters with 104 billion active, the cost to use the model has dropped by 33 percent. Input costs are now $2 per million tokens, down from $3, while output costs have fallen to $10 from $15. Despite the lower price, the model generates 14 percent more output tokens, meaning the cost per task remains almost identical to the previous version. This combination of lower prices and higher efficiency aims to make advanced AI more viable for everyday business applications.
Specific technical improvements are evident in specialized fields. In coding, the model scores 65.7 percent on Terminal-Bench 4.0, a test where its predecessor managed only 13 percent. This places it just behind Claude Opus 5.5 and ahead of GPT-6 Astra. In 3D modeling, the system can interact directly with game engines like Unity and Unreal Engine. It achieves a 92.7 percent score on the BenchCAD benchmark, nearly matching the performance of top proprietary models. Additionally, Moonshot has launched a feature called "3D Make" that allows users to preview and export 3D models directly through the interface.
Security considerations have also shaped the model's development. Moonshot observed that vulnerability research capabilities grew as post-training scaled up. To manage this, the company stripped cyber-related data from the training phases for the open-weights release. On an internal test reproducing known software vulnerabilities, the cleaned model solves 81.0 percent of tasks, compared to 90.5 percent for the uncleaned version. This approach balances the need for strong defensive capabilities with the desire to avoid pushing the absolute frontier of offensive cyber tools.
The release is available now through various platforms, including Kimi.ai and the Kimi API. A variant called K3.1 Pro Highspeed is also offered, which runs two to three times faster for users prioritizing speed over maximum depth. As these tools become more accessible and affordable, the distinction between premium proprietary systems and open alternatives continues to diminish, potentially reshaping how businesses and developers integrate AI into their workflows.
Source: The Decoder
