Dispatch
DeepSeek's refreshed V4.1 Flash beats its own V4 Pro preview on coding at roughly a third of the price
DeepSeek re-post-trained V4-Flash to Terminal-Bench 2.1 of 82.7 while holding $0.14/$0.28 per million tokens, and on MathArena's AIME 2026 set it scores 95.83% against V4 Pro's 96.67%, statistically indistinguishable at about a ninth of the cost per problem. New Flash pricing takes effect 10 September at $0.003 input cache hit, $0.15 cache miss and $0.60 output off-peak, with peak-hour rates double. Reported throughput is around 400 tokens/second, peaking near 427. The pattern worth watching is a lab's cheap tier overtaking its own flagship preview rather than a competitor's.
Source
↳ Follow the thread