Chinese AI Labs Are Racing Each Other Now, Not Just the West

Qwen 3.8 appeared with a banner announcement claiming 2.4 trillion parameters, positioning itself as second only to something called 'Fable' (likely a placeholder or internal model name). The thread immediately noted this was probably a response to Moonshot AI announcing Kimi K3 at 2.8T parameters, scheduled for Hugging Face release by July 27th. Two massive Chinese labs are now leapfrogging each other on parameter counts in a matter of days.

The pattern: the frontier AI race is no longer a US-vs-China story. It's fragmenting into multiple simultaneous competitions, with Chinese labs now racing each other on open weights releases. This is good for the ecosystem because it means more capable open models keep arriving, but it also makes the 'moat' of any closed model thinner by the week.

The comments were skeptical of raw parameter counts as a quality signal, which is fair, but the velocity of these releases is itself the story. When two labs are lobbing 2.4T and 2.8T parameter models at each other within days, the pace of capability improvement in open weights is accelerating.