What is Kimi K3, China’s first open AI model to reach 2.8 trillion parameters?
Chinese AI firm Moonshot AI announced its Kimi K3 model, an "open model" with 2.8 trillion parameters and a 1-million-token context window, claiming it's the world's first open 3T-class model. While it trails top US models like Claude Fable 5 and GPT 5.6 Sol, it outperforms older US rivals and is designed for frontier intelligence. US AI companies have raised concerns about Chinese firms releasing competitive models at lower rates and suspect data distillation.
Why it matters
This signifies China's advancement in frontier AI development, intensifying the global AI race and raising concerns among US firms about competitive practices and potential data "distillation" by Chinese companies, highlighting geopolitical tensions in AI innovation.
Chinese 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.
It also has a 1-million-token context window, with the company noting that Kimi K3 yielded about 2.5× improvement in overall scaling efficiency when compared to Kimi K2.
“Kimi K3 is a 2.8T-parameter model built on our Kimi Delta Attention and Attention Residuals, with native vision capabilities and a 1-million-token context window. It is the world’s first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning,” Moonshot AI said in a blog.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in