DeepSeek-V4, a new large language model, has officially launched its preview and open-source versions. Baidu AI Cloud's Qianfan platform has achieved Day0 adaptation, providing immediate API services for the model. A significant feature of DeepSeek-V4 is its support for an ultra-long context window of a million tokens, which is crucial for handling complex tasks requiring extensive background information. The model is available in two versions, DeepSeek-V4-Pro and DeepSeek-V4-Flash, designed to meet diverse scenario and performance requirements.
Currently, enterprise users and developers can directly access and integrate the DeepSeek-V4-Pro version through the Baidu Qianfan console or API. The DeepSeek-V4-Flash version is also set to be fully released soon, offering broader access and flexibility to users.