Why Chinese labs distill from Anthropic: Claude's agent data is the scarce training signal

teortaxesTex · x · 2026-09-11

Responding to a question on why Chinese labs seemingly distill from Anthropic rather than OpenAI, teortaxesTex argues OpenAI doesn't care much about China, but fundamentally it's about the difficulty of acquiring certain training signal: Claude is a known good agent, this is the era of agents, and agentic RL seed datasets are hard to get otherwise.

Related event: Why Chinese labs mostly distill from Anthropic, not OpenAI(3 posts)→

Original post →

More from Models

Models channel →