Chinese AI model cuts inference costs

Developers and businesses relying on large language models can expect lower running costs this quarter as a new Chinese entry undercuts established Western rivals. Moonshot AI has launched Kimi K3.1 Pro, a model that maintains high performance scores while reducing input and output token prices by one-third. This shift directly impacts operational budgets for teams integrating AI into their workflows, offering a more cost-effective alternative to leading proprietary systems.

The new model, released on October 23, 2026, jumps to fourth place on the Artificial Analysis Intelligence Index with a score of 55. It sits just three points behind the leading proprietary models, which include Anthropic's Claude Opus 5.5 and Sonnet 5.5. By narrowing the performance gap, Moonshot challenges the premium pricing previously commanded by top-tier Western providers. The release signals a tightening market where efficiency and cost are becoming as critical as raw capability.

For enterprise buyers, the economic structure of the new model is a key differentiator. Input costs drop to $2 per million tokens from $3, while output costs fall to $10 from $15. Cache hit prices also decrease to $0.20. Despite these lower rates, the total cost per task remains stable at approximately $1.98 because the model generates 14 percent more output tokens. This pricing strategy aims to make high-end reasoning accessible to a broader segment of the market without sacrificing quality.

Technical improvements are most visible in agentic tasks and coding. On the GDPval-AA benchmark, the model achieves an Elo score of 1,846, a significant increase from its predecessor. It also performs strongly in 3D modeling, working directly with game engines like Unity and Unreal Engine. These capabilities allow it to handle both quick fixes and full project builds, a feature that appeals to developers in the gaming and software sectors.

Moonshot also addresses cybersecurity concerns by removing specific data from its training phases before the open-weights release. The company aims to balance offensive and defensive capabilities, ensuring the model remains useful for auditing systems without pushing the absolute frontier of vulnerability discovery. This approach reflects a broader industry trend of carefully calibrating open models to manage risk while maintaining utility.

The model retains the same 2.8 trillion total parameter count as its predecessor, with 104 billion active parameters. It supports a context window of one million tokens and processes both text and images. A faster variant, K3.1 Pro Highspeed, is also available, running two to three times quicker for time-sensitive applications. Both versions are accessible via Kimi.ai, Kimi Work, and the API.

Market reaction will likely focus on the impending open-weights release scheduled for November 4, 2026. If realized, this would make Kimi K3.1 Pro the highest-scoring open-weights model available. This move could shrink the performance gap between open and proprietary models from 12 points to just three, potentially forcing competitors to adjust their pricing and release strategies in response.