Follow-up: a 3T-parameter model may already exist, scaling issues remain the wildcard

teortaxesTex · x · 2026-09-10

Following up on the extrapolation that V4.1 Pro is a 3.1T-total-param MoE, the author notes the caveat: this assumes they don't hit scaling issues again as with V4-Pro. The team sounds optimistic, and the author believes a 3T model likely already exists. Unverified speculation.

Related event: Rumor: DeepSeek Trained a 3T-Parameter Model but Held Back Release(3 posts)→

Original post →

More from Models

Models channel →