DeepSeek Releases DeepSeek-V4-Pro-0813, a New Model Registry Update on Hugging Face

DeepSeek updated its official Hugging Face model registry on August 13 with DeepSeek-V4-Pro-0813, a new entry in the DeepSeek-V4 family aimed at text generation.

According to the registry metadata, the model uses the transformers library and safetensors format, is tagged under the deepseek_v4 series, supports conversational tasks, and is marked endpoints_compatible, meaning it can be deployed through standard inference endpoints.

The model is released under the MIT license, a commercially friendly open-source license, and the registry entry links to arXiv paper 2606.19348. At the time of the update, the page showed zero downloads and 174 likes, a sign that this is a freshly published version.

The 0813 suffix in the model name points to a release dated August 13, and the Pro designation signals a strengthened variant in the V4 line, continuing DeepSeek's steady cadence of iterating on its flagship open-source models.

The launch matters because a Pro-tier release shows DeepSeek doubling down on both quality and deployment friendliness. The combination of an MIT license and the endpoints_compatible tag matters because it lowers friction for third-party platforms to plug the model into their own inference services.

The next things to watch are whether DeepSeek publishes technical details and benchmark results on the model card, and whether the official DeepSeek API begins serving the Pro variant.