At the World Artificial Intelligence Conference in Shanghai on Friday, Chinese startup Moonshot AI unveiled Kimi K3, a massive artificial intelligence model with 2.8 trillion parameters that the company claims can rival top American firms. The model, described by the firm as its "most capable flagship model to date," is set to be released as an open-source system on 27 July, making it the world's first freely available model in the three-trillion-parameter class. Its arrival comes just weeks after the US government forced American developer Anthropic to temporarily withdraw its flagship Fable and Mythos models over severe cybersecurity concerns, a move that highlighted Washington's view of advanced AI as critical national infrastructure subject to strict export controls.
Heavily backed by domestic tech giants Alibaba and Tencent, Moonshot has risen rapidly to the forefront of China's generative AI ecosystem. Kimi K3's open nature allows global users to modify it for advanced reasoning and complex software development, unlike closed proprietary systems from OpenAI or Anthropic. Independent evaluations from Artificial Analysis and Arena.ai show the model performing on par with leading US models such as OpenAI's GPT and Anthropic's Claude, with Kimi K3 ranking first in web interface engineering and outperforming Anthropic's Fable system in blind human-preference tests.
“China's Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter open-source model claiming to rival US firms.”
While the model's massive size requires significant computing equipment to run locally, its open-source release could disrupt Silicon Valley's commercial models. The announcement had an immediate impact on shares of Moonshot's domestic competitors Zhipu and MiniMax, which tumbled sharply in Hong Kong by about 27% and 16% respectively. The sudden breakthrough suggests that China's tech prowess is rapidly narrowing the capabilities gap, upending long-held Western assumptions that Chinese developers trail their American peers. Moonshot AI noted that the system is uniquely built to operate with "minimal human supervision" to sustain tasks such as engineering and coding.