DeepSeek-V4-Pro Final Version Launches with Full Open-Source Access

The official release of DeepSeek-V4-Pro-0813 is now live on HuggingFace. The complete model weights are available for download, allowing developers to immediately begin integration work. Notably, the project continues to use the permissive MIT license, facilitating unrestricted commercial use and distribution.

Immediate Availability

The finalized model weights are accessible on its HuggingFace repository. Alongside the release, the development team has published comprehensive documentation to streamline the setup process.

Official Deployment Pathways

To support diverse implementation needs, the official guides detail integration with two key inference frameworks:

  • vLLM Deployment: Recommended for production environments requiring high throughput and serving capabilities, known for its efficient memory management.
  • SGLang Integration: Offers optimizations for complex prompting and structured generation tasks, suitable for research and advanced applications.

Detailed instructions for local execution are also provided for developers testing on private hardware.

Impact on the Developer Ecosystem

This release delivers a stable model version coupled with robust tooling support, significantly reducing the barrier to practical AI application. The continued use of the MIT license is expected to drive widespread commercial adoption and innovation within the developer community.