ByteDance is training an AI model with up to 10 trillion parameters, approaching the scale of Anthropic's cutting-edge Mythos system.
That is more than 3x the size of Moonshot AI's Kimi K3 (2.8T params). Industry estimates put Mythos 5 at ~8T parameters and Fable 5 at ~5T. US labs don't disclose exact counts so comparisons are approximate.
Pre-training expected to take 3 to 6 months. Chinese AI labs continue narrowing the gap with top US companies.
Source: ft.com/content/9b8383b1-a28d-4…
