DeepSeek V4-Flash-0731 Beats Its Own Flagship on Nine Agentic Benchmarks at One-Third the Price

DeepSeek launched V4-Flash-0731 on July 31 via an API public beta, a retrained V4-Flash that outscores the larger V4-Pro-Preview across all nine agent and coding benchmarks while costing roughly one-third as much per output token. The architecture is unchanged, with gains coming entirely from improved post-training, including a 325 percent jump on the DeepSWE benchmark.