产品信号
DeepSeek V4 Flash走向完整部署
DeepSeek将V4 Flash API转为public beta,发布0731版本、agent基准、Codex支持、公开价格与MIT权重。
DeepSeek将DeepSeek-V4-Flash-0731 API推入public beta,调用仍使用deepseek-v4-flash,同时补齐公开价格、Responses API、Codex适配和更高并发。相较4月Preview,本次更新还公布更强的agent benchmark,并以MIT license提供权重,使模型可在托管API与自托管之间选择。
变化不只是一项发布标签
4月Preview阶段,V4 Flash已经可以通过API调用,支持1M context、OpenAI ChatCompletions与Anthropic API。此次0731版本维持model name不变,但DeepSeek称agent能力显著增强:Terminal Bench 2.1得分82.7、Cybergym为76.7、Toolathlon verified为70.3、DeepSWE为54.4。因而真正的增量是模型能力与部署接口同步升级,而非首次开放访问。
API与自托管形成双路径
当前API支持1M context、最高384K输出、JSON Output、Tool Calls、Responses API及Anthropic API。cache-miss input、output价格分别为每百万token 0.14美元和0.28美元,并发上限为2500,高于V4 Pro的500。DeepSeek还称现阶段只有V4 Flash支持Codex集成;与此同时,284B total、13B active的FP4与FP8 mixed权重已按MIT license开放下载和自托管。
竞争变量转向部署组合
这组变化把V4 Flash的竞争点从单一benchmark扩展到接口兼容、并发容量、token价格和部署控制权。Codex适配可以降低既有编码代理工作流的迁移成本,开放权重则给需要本地控制的团队另一条路径。最强反例是所有成绩均来自DeepSeek自身披露,且Preview此前已经可用;在独立复测与实际采用数据出现前,发布包的完整度不能直接等同于生产优势。
接下来关注什么
下一步应观察三项可核验信号:独立团队能否复现Terminal Bench 2.1、Cybergym和DeepSWE成绩;V4 Pro能否按计划在8月初支持Codex;以及开发者在2500并发下测得的延迟、稳定性和单位任务成本。这些结果将决定V4 Flash是接口完整的升级,还是能真正改变agent部署选择。
信源
- DeepSeek API Changelog — DeepSeek-V4-Flash Update
- DeepSeek API Docs — DeepSeek V4 Preview Release
- DeepSeek API Docs — Models & Pricing
- DeepSeek API Docs — Integrate with Codex
- Hugging Face / DeepSeek-AI — deepseek-ai/DeepSeek-V4-Flash
- Simon Willison — deepseek-ai/DeepSeek-V4-Flash-0731
- Product Hunt — DeepSeek-V4-Flash-0731
- DeepSeek — DeepSeek-V4-Flash-0731
- Artificial Analysis — DeepSeek V4 Flash (max) - Intelligence, Performance & Price Analysis
- The Information — DeepSeek Makes a Splash with Small, Affordable V4-Flash Model
- DeepSeek — Change Log
- DeepSeek API Changelog — DeepSeek-V4-Pro Update
- DeepSeek — Models & Pricing
- National Institute of Standards and Technology — CAISI Evaluation of DeepSeek V4 Pro
- The Decoder AI — Deepseek ships improved V4 Pro, open-sources its agent software, and raises API prices
- DeepSeek — deepseek-ai/deepseek-harness
- The Information — DeepSeek’s Flagship V4-Pro Model Gets Mixed Reviews
- DeepSeek — deepseek-ai/DeepSeek-V4-Pro
- Vals AI — DeepSeek V4
- Simon Willison — DeepSeek V4 Pro 0813 (on OpenRouter)
- DeepSeek — DeepSeek-V4-Pro GA Release
- DeepSeek — Your First API Call
- The Information — DeepSeek Announces Price Hikes for V4 Models
- Engadget — DeepSeek's AI models are about to cost four times more
- 腾讯云 — 〖大模型服务平台 TokenHub〗&〖智能体开发平台〗关于腾讯云 DeepSeek-V4 正式版〖原厂直供〗模型发布计划及计费调整通知
- DeepSeek API Changelog — DeepSeek-V4-Flash-Vision-Exp Release
- Terminal-Bench — How to run Terminal-Bench 2.1
- DeepSeek — Files API
- DeepSeek — DeepSeek-V4 Preview: Entering the Era of Affordable Million-Token Context
- Bloomberg Technology — DeepSeek Unveils Test Model to Rival Anthropic’s Opus 4.8
- DeepSeek — 更新日志
- DeepSeek — 模型 & 价格
- DeepSeek — 图像理解
- The Decoder AI — Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks
- Bloomberg Technology — DeepSeek Ends Weekend Peak Pricing for API Users From Today
- DeepSeek — 模型 & 价格 | DeepSeek API Docs
- TechNode — DeepSeek to introduce peak and off-peak pricing for its API
- 中华网 — DeepSeek再度调价 周末统一低谷价