Chinese AI start-up Moonshot has unveiled a vast new artificial intelligence model that it says can rival top US companies.
The company launched Kimi K3, which has 2.8 trillion parameters to measure the scale and processing power of AI.
The full capabilities of the Kimi K3 – coding, knowledge work and logic – will be known when it is released as an open-source model on July 27.
The sudden success shows that China’s technological prowess is rapidly closing the capabilities gap, shattering long-held perceptions in the West that Chinese developers lag behind their American counterparts.
Its arrival later this month will make it the world’s first open-source model in the three-trillion-parameter class that can be freely downloaded, run and adapted by external developers.
The release comes at a highly sensitive moment for the global technology sector, just weeks after the US government suddenly forced American developer Anthropic to temporarily withdraw its flagship Fable and Mythos models due to serious cybersecurity concerns.
While Washington has since lifted those restrictions, the initial move highlights how the U.S. government now views advanced AI software as critical national infrastructure, labeling marginal models as critical national security assets subject to strict export controls.
However, the rapid arrival of the Kimi K3 shows that Chinese companies are successfully overcoming these regulatory hurdles and moving forward independently despite US restrictions on hardware sales.
With heavy backing by domestic tech giants Alibaba and Tencent, Moonshot has quickly risen to the forefront of China’s generative AI ecosystem.
The K3 is Moonshot AI’s “most capable flagship model yet,” the company said in a statement.
Unlike the closed, proprietary US systems of OpenAI or Anthropic, the open nature of Kimi K3 allows global users to modify the system for advanced logic and complex software development.
Moonshot AI noted that the system is uniquely designed to operate with “minimal human supervision” to maintain tasks such as engineering and coding.
Artificial analysis and third-party evaluation of Arena.ai shows that the model is performing on par with leading models in the US such as OpenAI’s GPT and Anthropic’s Cloud.
In these independent benchmarks, externalKimi K3 takes first place in web interface engineering, outperforming Anthropic’s Fable system, external In blind human-preference tests.
While the sheer size of the system means it requires significant computing equipment to run locally, making it open-source could drastically disrupt Silicon Valley’s commercial model.
The announcement had an immediate impact on shares of Moonshot’s domestic competitors Zipu and Minimax, which fell sharply by about 27% and 16% respectively in Hong Kong.