ByteDance Reportedly Pre-Training Toward a 10-Trillion-Parameter Model

As Chinese labs push model scale into new territory and raise API prices, US vendors are cutting theirs — a divergence that says as much about strategy as it does about compute.

ByteDance is reportedly pre-training toward a 10-trillion-parameter model, according to reports surfaced by both @ivke2006 and @TraffAlex. If the figure holds, it would place ByteDance's next system at a parameter count that closes — at least on paper — the gap with the largest frontier systems built in the US.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.