Moonshot AI, a Beijing startup backed by Alibaba and Tencent, just handed out the complete blueprint of its newest AI system, called Kimi K3, for free. Anyone with enough computing power can now download it, copy it, and run it on their own servers, instead of only being allowed to rent access through an app the way you use ChatGPT or Claude.
This is a big model. It has 2.8 trillion internal settings, more than any other open model released so far, and it can process documents up to a million tokens long in a single conversation, roughly the length of several long novels. Moonshot says a new internal design lets the model squeeze more intelligence out of the same computing budget, about two and a half times more than its previous version. What matters more for outsiders is that on tests built to mimic actual office work, filling out spreadsheets, browsing the web to finish a task, running multi-step workflows, Kimi K3 matched or beat the newest paid models from OpenAI and Anthropic.
This is not the first time a Chinese lab has done this. In January 2025, a company called DeepSeek released a similarly capable model at a much lower cost, and the shock wiped billions off the market value of chipmakers like Nvidia within a single week. Kimi K3 triggered the same reaction: semiconductor stocks fell sharply again in the days after release, as investors worried that if a free, efficient model can do the job, companies will not need to keep buying as many expensive chips.
For most businesses, actually downloading and running Kimi K3 yourselves is not realistic. A model this size needs a room full of top tier server chips worth well over a hundred thousand dollars, which only cloud companies and a handful of large enterprises can afford. But that is not the point for most readers. The real effect is on price. Moonshot charges a fraction of what OpenAI and Anthropic charge for similar capability through its own hosted service, and that forces every AI vendor, including the big US names, to justify their prices or cut them.
Moonshot's rise has been fast. The company's valuation jumped from around four billion dollars at the end of last year to somewhere between twenty and fifty billion dollars now, and it is preparing to list on the Hong Kong stock exchange within months. That timeline suggests this release was not just a research showcase, it was designed to prove the company deserves that valuation before it goes public.
The practical takeaway for any business buying AI services is simple: do not assume today's AI subscription price is fixed. Competition from Chinese open models is real, it is speeding up, and it gives you leverage the next time you negotiate a contract with an AI vendor, even if you never touch the raw model files yourself.