News

The official version of DeepSeek V4 Pro is online: Agent capability upgrade API enables peak and valley pricing DeepSeek today announced that the official version of DeepSeek V4 Pro is online. It has been updated to the APP, web page and API simultaneously.

2 min read
The official version of DeepSeek V4 Pro is online: Agent capability upgrade API enables peak and valley pricing. DeepSeek today announced that the official version of DeepSeek V4 Pro is online. It has been simultaneously updated to the APP, web page and API. Users can use the official version of V4 Pro through "expert mode" on the APP and web pages, and the API model name remains unchanged. According to DeepSeek, this official version focuses on enhancing Agent capabilities, especially performance optimization for the production environment. For the Code Agent task in the public benchmark test set, DeepSeek-V4-Pro-0813 uses the DeepSeek Harness minimalist mode for testing. The test settings are max gear, topp=0.95, and temperature=1.0. DeepSeek also reminds that the test results may be different under other frameworks. In terms of API, the DeepSeek API now natively supports the OpenAI Responses API format and has been adapted to the Codex. Developers can complete Codex configuration through the official one-click configuration script. At the same time, the thinking modes of V4-Pro and V4-Flash have added different intensity settings, supporting three levels of low, high and max. Among them, low is mainly for simple tasks, high is for daily agent tasks, and max is for complex task scenarios. Developers can adjust the corresponding gear according to the complexity of the task. As the official version of the DeepSeek V4 series goes online, its API price will also be adjusted. DeepSeek announced the adoption of a peak and valley pricing mechanism. The price during off-peak hours is half of that during peak hours. The new price will officially take effect at 0:00 on August 17th, Beijing time. via cnBeta.COM - Chinese industry information website (author: Source: Phoenix Technology)