Moonshot AI cuts inference costs as Kimi K3.1 Pro narrows gap with US rivals

Moonshot AI has released Kimi K3.1 Pro, a post-trained model that has surged to the fourth position on the Artificial Analysis Intelligence Index. The update marks a significant shift in the competitive landscape, placing the Chinese firm within striking distance of the leading proprietary models from Anthropic and others. By combining substantial performance gains with reduced pricing, Moonshot is positioning its latest offering as a highly competitive option for enterprises seeking to balance capability with operational expenditure.

The financial implications of the release are immediate. Moonshot has reduced API token prices by 33% compared to its predecessor. Input costs now stand at $2 per million tokens, down from $3, while output costs have fallen to $10 from $15. Cache hit prices have also been lowered to $0.20 per million tokens. Despite these price cuts, the cost per Intelligence Index task remains nearly identical to the previous version at $1.98, a figure that is stable because the new model generates 14% more output tokens. The model retains the same architecture, featuring 2.8 trillion total parameters with 104 billion active, and supports a context window of one million tokens for both text and image inputs.

Performance metrics indicate a strong showing in specialized domains. On the Artificial Analysis Intelligence Index, Kimi K3.1 Pro scored 55 points, an 11-point improvement over the K3 version. This places it just three points behind Claude Opus 5.5, which leads the proprietary field with 58 points. The only models ranking higher are Claude Sonnet 5.5 and Muse Wonder Full, both at 56 points. In agentic tasks, the new model demonstrates particular strength, achieving a GDPval-AA Elo of 1,846, which is only 20 points below the leader. On Terminal-Bench 4.0, it recorded a success rate of 65.7%, narrowly trailing the top performer at 66.4%.

The company is also targeting the 3D modeling and coding markets. Kimi K3.1 Pro integrates directly with game engines such as Unity and Unreal Engine, and scores 92.7% on the BenchCAD benchmark when using tools. Moonshot has also launched a '3D Make' feature on its platform that allows users to preview and export models. To address security concerns regarding the model's offensive cyber capabilities, Moonshot removed cyber-related data from the training set for the open-weights version. This adjustment reduces its ability to solve specific vulnerability tasks from 90.5% to 81.0%, a figure that remains higher than the previous K3 model but slightly below DeepSeek-V4.1-Flash.

Strategic timing will determine the model's long-term impact on the open-weights sector. Moonshot plans to release the model weights by November 4, 2026. If executed as scheduled, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This move would shrink the performance gap between the best open models and proprietary leaders from 12 points to just three, assuming no other major releases occur in the interim. The company also offers a 'Highspeed' variant that runs two to three times faster, expanding the utility of the platform for developers requiring lower latency. All variants are currently accessible via Kimi.ai, Kimi Work, Kimi Code, and the public API.