DeepSeek releases V4.1-Flash: 552B MoE claimed to beat V4-Pro on cost and speed

eyishazyer · x · 2026-09-11

DeepSeek officially released V4.1-Flash on Sept. 10: a 552B-parameter MoE with a new causal encoder-decoder split (8B active params for input, 16B for output). DeepSeek claims its KV cache uses a quarter of the HBM and an eighth of the SSD footprint of the previous generation, and that Flash beats V4-Pro on performance, cost, speed and runtime — enough that all V4-Pro API traffic will be routed to Flash from Sept. 14. Off-peak cache-hit input pricing is reportedly dropping to RMB 0.02 per million tokens, per TechNode, though unconfirmed on DeepSeek's pricing page.

Related event: DeepSeek Open-Sources V4.1 Flash: 552B MoE New Architecture Beats Its Own Flagship at a Fraction of the Cost(52 posts)→

Original post →

More from Infra

Infra channel →