Yang Zhilin

“now reportedly raising at a $50 billion valuation”

“Kimi K3 was officially released on July 16 and showed itself seriously competitive with American frontier models on performance benchmarks and price...it's about one-third the cost of OpenAI, Anthropic, or Gemini 3.1 Pro.”

“Moonshot's specific edge: switching from Adam to the Muon optimizer, which accounts for dependencies between parameters rather than treating them independently. In effect, 30 trillion tokens behaves like 60 trillion — a 2x efficiency gain with no additional data.”

“The market conflates benchmark reasoning scores with real-world agent performance — but they are built differently and scale differently. Yang argues that acting against live environments is its own capability axis.”

“Cloud's reasoning performance is not very high, but its performance as an agent is very high... The contrarian implication: benchmark leaderboards are the wrong scorecard for product investors. A model that ranks 4th on reasoning but 1st on real-world task completion is more valuable as a product.”

“He compares it to the steam engine... the same lag Demis Hassabis maps onto the next few years.”

“He had a very clear Tech Vision. On Day 1, he spoke about building a Super App. And this was when they didn't even have a model yet, let alone an app.”

“Wang Xinyu led Longzhu's investment in July 2023 — the fund's only LLM investment.”

“Luo Tianchen's generational vintage led him to solve the hardest AI problem of his era — L4 autonomous driving problems within computer vision. Yang Zhilin's vintage led him to solve large language models in NLP.”