DeepSeek-V4-Flash 官方发版：Agent 能力大幅提升，原生支持 Responses API
Source: <https://api-docs.deepseek.com/updates> | 2026-07-31
Author: DeepSeek | Category: models | Quality Score: 4/5
Tags: deepseek, llm, agent, benchmark, api
English Summary
DeepSeek officially released the V4-Flash API into public beta on July 31, 2026. The model keeps the same architecture and size as V4-Flash-Preview but was re-post-trained, with agent capabilities significantly surpassing V4-Pro-Preview. Key benchmarks: Terminal Bench 2.1 = 82.7, NL2Repo = 54.2, Cybergym = 76.7, DeepSWE = 54.4, Toolathlon Verified = 70.3, Agent Last Exam = 25.2. Natively supports the Responses API format and is specifically adapted for Codex. The official V4-Pro release will follow soon.
中文概要
DeepSeek 于 7月31日官方发布 V4-Flash API 公测版该版本与 V4-Flash-Preview 保持相同架构和规模，仅进行了重新后训练，但 Agent 能力全面超越 V4-Pro-Preview基准成绩：Terminal Bench 2.1 = 82.7NL2Repo = 54.2Cybergym = 76.7DeepSWE = 54.4Toolathlon Verified = 70.3Agent Last Exam = 25.2DSBench-FullStack = 68.7同时原生支持 Responses API 格式并专门适配 CodexV4-Pro 官方版将于稍后发布