What's even wilder about GLM-5.2 is its parameter size.
Opus and GPT are estimated to be around 2T, and Mythos/Fable are several times larger than that.
But GLM matches or beats their performance at just 750B, 1/3 of the size. And it's open-weight.
I don't know how much longer the fake narrative marketing against Chinese AI models will hold up, but the industry is definitely in a state of emergency.
Over 70% of SF startups, including big players like Uber, already pivoted to open-source AI earlier this year.
Now, the gap has completely vanished.