AI BriefingDeepSeek-V4-Flash Update
Key point
DeepSeek has released the V4-Flash API in public beta with enhanced agent performance.
Details
The official DeepSeek-V4-Flash API has entered public beta. The existing API call method remains unchanged, and specifying deepseek-v4-flash as the model name will use the latest version.
Agent performance significantly exceeded V4-Pro-Preview.
- 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
- DSBench-Hard: 59.6
The official V4-Flash supports the Responses API format by default and has been tuned for Codex. For Code Agent tasks in the public benchmarks, it was tested using the soon-to-be-released DeepSeek Harness minimal mode, with max effort, topp=0.95, and temperature 1.0 settings.
DeepSeek-V4-Flash-0731 retains the same model architecture and scale as V4-Flash-Preview, having undergone only additional training. This update applies only to the V4-Flash API, and the V4-Pro API as well as the APP/WEB models remain unchanged.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen Intelligence Launches Three Mobile AI Agents and Open Benchmark Suite