DeepSeek Launches V4.1-Flash as Cheaper Multimodal Model for Agent Workloads
The open-weight MoE model uses a new Causal Encoder–Decoder design as DeepSeek prepares to route V4-Pro API traffic to Flash.
DeepSeek
DeepSeek Launches V4-Flash API With Major Agent Upgrade, Matching Opus 4.8 at a Fraction of the Cost
The upgraded API dramatically improves coding and agent benchmarks without changing the underlying model, underscoring how inference efficiency is becoming a new competitive battleground.