Moonshot AI releases Kimi K3, a 2.8T parameter MoE model with open weights scheduled for July 27

Moonshot AI released Kimi K3, a 2.8 trillion parameter Mixture-of-Experts model, with its weights scheduled for public release on July 27, 2026. The model is described as the closest open model to the frontier since DeepSeek R1 and is positioned as the strongest open model released to date.

Kimi K3 ranks second overall on the Vals AI index, third on Artificial Analysis’s Intelligence Index—beaten only by Claude Fable and GPT-5.6 Sol Max while being cheaper—and first on Frontend Code Arena. The model’s performance places it alongside leading closed models from Anthropic and OpenAI despite Moonshot AI operating with significantly fewer resources.

The release of Kimi K3 is framed as evidence that Chinese AI labs can achieve frontier performance through execution on scaling, data, algorithms, architecture, tools, and environments, rather than relying primarily on adversarial distillation from U.S. frontier models.

The analysis notes that China’s government recently signaled a commitment to open-source AI at the World AI Conference, with Xi Jinping’s keynote address emphasizing global diffusion of AI technologies. This commitment coincides with the announcement of Kimi K3 and suggests a strategic tolerance for open-weight releases.

The report also highlights a shift in the balance between open and closed models, with open models increasingly competitive in performance while maintaining economic advantages in distribution and adoption.