DeepSeek Launches Public Beta API for DeepSeek-V4-Flash with Enhanced Agent Capabilities
DeepSeek has launched the public beta API for the official version of DeepSeek-V4-Flash, featuring significantly upgraded agent capabilities. The new model outperforms the previous V4-Pro-Preview across multiple agent-focused benchmarks, including Terminal Bench 2.1 and DSBench. This release demonstrates that targeted post-training can dramatically boost agentic performance without increasing model size or computational overhead. It provides developers with a faster, more cost-effective "Flash" model that is highly capable of handling complex, autonomous tasks. The model structure and size of DeepSeek-V4-Flash-0731 remain identical to the preview version, as the improvements stem entirely from post-training. Additionally, it natively supports the Responses API format and is optimized for Codex integration.
## BACKGROUND
AI agents are systems powered by LLMs that can autonomously plan, use tools, and execute multi-step tasks in environments like terminal consoles or code repositories. Benchmarks like Terminal Bench (for command-line tasks), NL2Repo (for repository-level code generation), and DSBench (for data science) are specifically designed to measure these agentic capabilities.