Yang Zhilin turned down an Apple job reporting into Tim Cook's inner circle, and Moonshot AI, the company he built instead, just released a 2.8 trillion parameter model for free. Key Points:
Yang Zhilin turned down an Apple job reporting into Tim Cook's inner circle, and Moonshot AI, the company he built instead, just released a 2.8 trillion parameter model for free.
- Moonshot AI put the full weights of Kimi K3 into public hands on Jul. 27, making a 2.8 trillion parameter model free to download, modify and self-host.
- Founder Yang Zhilin, 34, rejected an Apple offer at the end of his Carnegie Mellon doctorate and returned to Beijing to start his own lab.
- The release has pulled Washington into an open fight over open weight models, distillation claims and chip export controls.
Moonshot AIpublished the full weights of its Kimi K3 model on Jul. 27, letting developers download, adapt and run it on their own hardware.
The Beijing lab had served the model through its API and consumer app since Jul. 16, holding the weights back for ten days.
At 2.8 trillion parameters, the release ranks as the largest open weight system ever handed to the public, and the download runs to roughly 594 gigabytes in its native format.
The model fires only a fraction of those parameters on each token, which keeps the running cost closer to that of a mid sized system. It reads up to one million tokens in a single pass. Demand after the debut climbed high enough that Moonshot capped new subscriptions.
Pricing tells its own story, with K3 costing $15 per million output tokens against $50 for Anthropic's Fable 5. Earlier Kimi models already ran inside American products, including a coding agent built by Cursor, so the customer base existed well before K3 arrived.
Also Read:Crypto Wrench Attacks Reach 52 Cases In Six Months, Most Of Them In France
Yang, 34, was born in Shantou, Guangdong, studied computer science at Tsinghua University, then finished a doctorate at Carnegie Mellon in four years. His adviser there was Ruslan Salakhutdinov, a machine learning professor who has held AI roles at both Apple and Meta. Salakhutdinov wrote on X last week that he wanted to clear up the confusion about why his former student left.
Apple pushed hard to hire him, according to that account, and the company was willing to base him in its Beijing office rather than California.
A senior executive reporting directly to Cook wanted him on the team. Yang said no.
Salakhutdinov recalled him saying that he would regret it for the rest of his life if he never tried building his own company.
Mohammed Soliman, a director at the Washington advisory firm McLarty Associates, said K3 is not necessarily better than American models, and that the point may not matter much. It is close enough for the purpose, he argued, even with the semiconductor limits China works under. "The gap has closed faster than a lot of people expected," he said.
White House science and technology policy director Michael Kratsios said last week that the government holds evidence Moonshot improperly accessed American models, and Treasury Secretary Scott Bessent floated sanctions days later. Moonshot rejects the charge.
Anthropic chief executive Dario Amodeianswered on Monday.
His company has never called for a ban on open weight models, he wrote, and models without dangerous capabilities serve as a public good. He asked instead for tighter chip controls, enforcement against industrial scale distillation, and safety testing for every capable model regardless of origin.
Moonshot's ascent has been fast even by Chinese standards, and the firm raised $2 billion in May at a valuation above $20 billion before reaching $300 million in annual recurring revenue in June. It is now courting investors at $50 billion ahead of a possible Hong Kong listing, three years after Yang walked away from Cupertino.
Read Next:XRP Near-Term Forecasts Top Out At $1.25 Before The Fed Decision
