Yang Zhilin Moonshot K3

par Vojtech: Yang Zhilin

Everyone assumed Yang Zhilin was too late to the AI race after studying at Carnegie Mellon and deciding to leave the US for China instead. Professors were genuinely surprised by that choice, but he started Moonshot anyway. The strategy wasn't about chasing every shiny use case because "if you try to do everything, you do nothing well." They focused entirely on scaling laws as the equivalent of Moore's Law for AI to prioritize productivity. Kimi launched and kept breaking open-source records repeatedly. Then in July 2026, they released model K3 with 2.8 trillion parameters, described as the largest open model ever created. It costs one-third of the price of frontier models while the company hit a valuation of $50 billion. Despite all that, Yang Zhilin refuses to call it a success yet, characterizing the situation as very early PMF. He is described as running a marathon while everyone else is sprinting.