On June 18, Elon replied to a tweet saying that Chinese models would catch up to US models, especially Mythos- and Fable 5-level capabilities, by late this year. But Elon replied, "Probably Q1," meaning early 2027. Then a Chinese researcher, Jietang from GLM Labs, replied, "Won't take that long."

So yes, everyone thought it would happen through one of the big Chinese labs like GLM, DeepSeek, or maybe someone else. But no one was ready for how fast it would happen.

Literally less than a month later, Moonshot AI, a Chinese AI lab, launched a 2.8T-parameter monster model built on Kimi Delta Attention and Attention Residuals, with native vision capabilities and a 1-million-token context window.