DeepSeek has released V4.1-Flash, a lightweight model it says reaches flagship-level capability at lower cost — a reversal of the pricing direction the company had signaled.
DeepSeek released V4.1-Flash, describing it as smarter, faster and more efficient. The company positions it as a lightweight tier that nonetheless matches its own flagship. Japanese coverage framed the release around the reversal: after signaling a price increase, DeepSeek is now marketing this model on lower cost rather than a higher one.
What moved here is less the model than the pricing assumption behind it. If a "Flash" tier can carry flagship-level work, the case for keeping an expensive top tier expensive weakens, and the competitive question shifts from how capable a model is toward how cheaply inference can be sold. The performance claims rest on DeepSeek's own account for now; independent evaluation of V4.1-Flash has not yet appeared.
The published pricing, and whether the top tier is repriced alongside it. A discount confined to Flash reads as a promotional hook; a change that reaches the flagship reads as a shift in strategy.