What is Kimi K3, China’s first open AI model to reach 2.8 trillion parameters?

Moonshot AI said that Kimi K3 was still behind models such as Claude Fable 5 and GPT 5.6 Sol, but hailed its capacity for coding, knowledge work, and multi-media creation

Moonshot AI said that Kimi K3 was still behind models such as Claude Fable 5 and GPT 5.6 Sol, but hailed its capacity for coding, knowledge work, and multi-media creationChinese AI firm Moonshot AI announced its Kimi K3 model, which it stressed was the first open model (not open-source model) to reach 2.8 trillion parameters.

The startup acknowledged that the overall performance of Kimi K3 trailed that of “powerful proprietary models” from the U.S. such as Claude Fable 5 and GPT 5.6 Sol, but that it delivered frontier-level performance across its evaluation suite and outperformed older rival models from the U.S. across several benchmarks.

American AI companies and technologists have raised concerns about Chinese AI firms releasing competitive frontier models at what they claim are far cheaper rates. Several U.S. firms also suspect that Chinese AI companies are “distilling” their data to build AI models at lower costs and draw in more users.

Anthropic claimed that it had identified industrial-scale campaigns by DeepSeek, Moonshot, and MiniMax to “illicitly extract Claude’s capabilities to improve their own models.”