When do Chinese models surpass US models? I thought there was at least be a 2 year runway but now I think they surpass it within 12 months, if not sooner.

I get the impression that they're focusing more on efficiency than raw intelligence. While I assume that all AI labs realize that the AGI "race" is mostly bs, the Chinese labs aren't stuck in a trap where they need to keep blowing money to maintain an intelligence lead to justify investments and valuations.

So, while OAI/Ant have to fearmonger and keep training the largest models, Chinese labs can focus on efficiency more heavily and as long as they stay near frontier, they'll continue to get positive coverage.