China's Moonshot releases its breakthrough AI model Kimi K3 for download
Today, Moonshot AI will make Kimi K3 — an AI model that has already created a buzz around itself — available for public download.
Today, Moonshot AI will make Kimi K3 — an AI model that has already created a buzz around itself — available for public download.
Today, Moonshot AI will make Kimi K3 — an AI model that has already created a buzz around itself — available for public download.
On Monday, July 27, the Beijing-based company plans to make the Kimi K3 scales — the parameters that guide the work of artificial intelligence when forming answers — publicly available, Bloomberg reports .
This will allow developers to freely download, customize, and deploy the technology on their own, which will help Moonshot expand its audience. Company founder Yang Zhilin said that he aims to gain user support through openness and wider access than closed American counterparts.
Like DeepSeek's early 2025 launch , K3's debut was seen as proof that Chinese companies can keep up with the times even with smaller budgets and more limited access to cutting-edge resources than their competitors with unfettered access to Nvidia Corp.'s AI accelerators.
Recall that Moonshot developed K3 with 2.8 trillion parameters , making it the world’s largest model with open weights, while leading competitors such as Anthropic keep the size of their models (like Claude) secret. K3 also has a context window of 1 million tokens, allowing it to process massive documents or code bases in a single query. While the raw weights of the model alone provide little human-readable information, Moonshot is preparing to reveal more details about its architecture, training, and test benchmarks in an upcoming white paper.
Moonshot’s success has put it at the center of a technology standoff between the U.S. and China. Last week, a White House official accused the company of training K3 on banned Nvidia chips, and also renewed accusations of distillation — the practice of unsanctionedly extracting answers from leading competitors’ models to improve its own performance. Moonshot did not immediately respond to requests for comment.



