Chinese artificial-intelligence start-up Moonshot has unveiled a massive new model it says can rival the leading American developers. The company launched Kimi K3, a system containing 2.8 trillion parameters, a figure that serves as a rough measure of an AI's scale and processing power.
The world's first open-source model in the three-trillion-parameter class — free to download, run and customise.
Kimi K3's full capabilities, spanning coding, knowledge work and reasoning, will become clear when it is released as an open-source model on 27 July. Its arrival will make it the world's first open-source model in the three-trillion-parameter class that can be freely downloaded, run and customised by outside developers, a milestone that suggests China's technological prowess is rapidly narrowing the gap with the West.
The breakthrough upends long-held assumptions that Chinese developers trail their American peers. Heavily backed by domestic giants Alibaba and Tencent, Moonshot has quickly risen to the forefront of China's generative-AI ecosystem. In a statement the company called K3 its "most capable flagship model to date."
The release lands at a delicate time for the global technology sector, just weeks after the US government abruptly forced American developer Anthropic to temporarily withdraw its flagship Fable and Mythos models over severe cybersecurity concerns. Washington has since lifted those restrictions, but the episode highlighted how the US now treats advanced AI as critical national infrastructure subject to strict export controls.
Yet the rapid arrival of Kimi K3 suggests Chinese firms are successfully bypassing those barriers and advancing independently despite curbs on hardware sales. Unlike closed American systems from OpenAI or Anthropic, Kimi K3's open nature lets global users modify it for advanced reasoning and complex software development, and Moonshot says it is built to operate with "minimal human supervision" on tasks such as engineering.
Third-party evaluations from Artificial Analysis and Arena.ai show the model performing on a par with leading US systems such as OpenAI's GPT and Anthropic's Claude. In independent blind human-preference tests, Kimi K3 ranked first in web-interface engineering, outperforming Anthropic's Fable. While its size means running it locally demands significant computing power, going open source could heavily disrupt Silicon Valley's commercial models.
The 27 July release now functions as a deadline for the industry's assumptions: after it, anyone will be able to test whether the gap between East and West has truly closed.
The parameter count is marketing; the licence is strategy. An open-weights model at this scale cannot be un-released, sanctioned, or export-controlled, which makes it a direct answer to Washington's treatment of AI as national infrastructure. If K3's benchmarks survive independent scrutiny after 27 July, the pricing power of closed US labs — not their capability lead — is what comes under pressure first.
