News
The official version of DeepSeek-V4-Pro is online Today, we release the official version of DeepSeek V4 Pro
2 min read
Source: Telegram AI频道
DeepSeek-V4-Pro official version is online Today, we release the official version of DeepSeek V4 Pro. Currently, the official version of DeepSeek V4 Pro has been launched simultaneously on the APP, web page and API updates. Users can select "Expert Mode" through the APP or web page to use the new V4 Pro official version model. The API model name remains unchanged. For other update details, please refer to the DeepSeek API official interface document. ----------------------- Agent capabilities have been greatly improved. The official version of DeepSeek V4 Pro has enhanced Agent capabilities, and the performance improvement in the production environment is particularly significant. *For the Code Agent task in the public benchmark test set, DeepSeek-V4-Pro-0813 uses the DeepSeek Harness minimalist mode as the framework for testing (using the max gear, topp=0.95, temperature=1.0), and the results under other frameworks may be slightly different. ----------------------- API support Native support for Responses API DeepSeek API now natively supports OpenAI Responses API format and is specifically adapted to Codex. Users can refer to the official API documentation: Quick Start - Access Agent Tool - Codex, and complete Codex configuration through one-click configuration script. More flexible thinking intensity control V4-Pro and V4-Flash thinking modes now support three levels of thinking intensity: low / high / max. Users can flexibly choose according to task complexity in actual use: use low for simple tasks, high for daily Agent tasks, and max for highly complex task scenarios. For the setting method, please refer to the official API documentation: API Guide - Thinking Mode. ----------------------- API pricing adjustment With the official release of the full series of DeepSeek V4 models, we will update and adjust the API price. In order to allocate resources more reasonably, we will adopt peak and valley pricing. The price during idle times is half of the price during peak hours, and users are encouraged to adjust task time according to actual usage. The new price will be effective at 0:00 Beijing time on August 17, 2026