DeepSeek-V4-Flash API launches in public beta with major agent upgrades
DeepSeek officially releases the V4-Flash API in public beta with a re-post-trained 0731 checkpoint, big agent benchmark gains, and native Responses API support.
BY FOUNDERBUILT AI NEWS
DeepSeek has released the official DeepSeek-V4-Flash API in public beta, making the latest version available to all developers through the existing endpoint. The calling method is unchanged: simply set the model name to deepseek-v4-flash. The update arrives with a re-post-trained checkpoint, V4-Flash-0731, which keeps the same model architecture and size as the earlier preview while delivering substantially stronger agent performance.
The new release shows big gains across agent benchmarks, far exceeding V4-Pro-Preview. DeepSeek reports Terminal Bench 2.1 at 82.7, NL2Repo at 54.2, Cybergym at 76.7, DeepSWE at 54.4, Toolathlon verified at 70.3, and Automation Bench at 25.1, alongside strong scores on its internal DSBench full-stack and hard problem sets (68.7 and 59.6). Public benchmark agent tasks were run with the DeepSeek Harness minimal mode at max effort, with topp 0.95 and temperature 1.0.
DeepSeek-V4-Flash also natively supports the Responses API format and is specifically adapted for Codex, positioning it as a strong coding-agent model. The update only affects the V4-Flash API: the V4-Pro API and the app and web models are unchanged. The official release of DeepSeek-V4-Pro is expected to follow soon.