DeepSeek V4-Flash API Released with Enhanced Agent Benchmarks

DeepSeek launches V4-Flash API featuring improved coding and tool use, surpassing V4-Pro-Preview in agent benchmarks without structural model changes.

Woofun AI reports that DeepSeek has officially released the V4-Flash API, prioritizing enhancements in programming, tool invocation, and long-task execution. The update natively supports the Responses API and includes Codex adaptation, while maintaining the existing model structure and parameter scale through focused retraining. Official statements indicate that V4-Flash achieves significantly higher scores on multiple Agent benchmarks compared to the previous V4-Pro-Preview. This release applies exclusively to the API interface, leaving the DeepSeek App and web versions unchanged, with the official V4-Pro version expected to launch shortly.