DeepSeek-V4-Flash API enters public beta
Original: DeepSeek-V4-Flash Update
Why This Matters
DeepSeek's V4-Flash public beta signals rapid iteration in agentic coding capabilities, intensifying competition in the AI model API market.
DeepSeek released DeepSeek-V4-Flash as a public beta API on July 31, 2026. The model retains the same architecture as V4-Flash-Preview but underwent re-post-training, significantly boosting agent capabilities with benchmark scores such as Terminal Bench 2.1 at 82.7 and DeepSWE at 54.4.
DeepSeek announced on July 31, 2026 that DeepSeek-V4-Flash is now officially in public beta via its API. Users can access the model by setting the model name to 'deepseek-v4-flash' — no changes to the API calling method are required. The update focuses on significantly enhanced agent capabilities, with benchmark scores reported as follows: Terminal Bench 2.1 (82.7), NL2Repo (54.2), Cybergym (76.7), DeepSWE (54.4), Toolathlon verified (70.3), Agent Last Exam (25.2), Automation Bench Public (25.1), DSBench-FullStack (68.7), and DSBench-Hard (59.6). DeepSeek-V4-Flash-0731 shares the same model architecture and size as DeepSeek-V4-Flash-Preview and was only re-post-trained. The model natively supports the Responses API format and is specifically adapted for Codex. Benchmark tests used DeepSeek Harness minimal mode with max effort, topp=0.95, and temperature=1.0. This update applies only to the V4-Flash API; the V4-Pro API and APP/WEB models remain unchanged. DeepSeek stated that the official release of DeepSeek-V4-Pro will follow soon.