DeepSeek Releases V4 Pro Open Weight Model With MIT License on Hugging Face

DeepSeek has announced the availability of its latest model, DeepSeek V4 Pro 0813, on the Hugging Face platform under an MIT license with open weight status. The announcement was first made via the X account @eliebakouch and subsequently retweeted by the official Hugging Face account, signaling validation from the model distribution platform. The model's naming has drawn some criticism from observers, who consider the "V4" designation somewhat misleading. The previous version was only a preview, while this latest release has undergone far more extensive training. According to sources, a more accurate comparison would be that this model feels like a leap to version 4.5 rather than a minor update from V4.

The official model page on Hugging Face, accessible at deepseek ai/DeepSeek V4 Pro, features the company's tagline about its mission to advance and democratize artificial intelligence through open source and open science. This move underscores DeepSeek's commitment to an open approach in AI development, a stance consistent with its previous releases. Interestingly, the model also appears on another Hugging Face account under the name DeidaraDemi/DeepSeek V4 Pro. However, the primary confirmed source is the official deepseek ai repository, while the second account is likely a mirror or re upload by a community user. DeepSeek's decision to use the MIT license sends a strong signal to the developer ecosystem. This license is among the most permissive in the software world, allowing commercial use, modification, and redistribution with minimal restrictions.

For companies and independent developers, this means the model can be integrated into commercial products without the obligation to share derivative code. Open weight models mean that the trained model weights are available for direct download and use, unlike closed models that are only accessible via API. This approach offers greater flexibility for developers who wish to fine tune the model or run it on their own infrastructure. The launch also comes with anticipation for the release of DeepSeek Harness, a tool mentioned in the announcement and greeted with enthusiasm by the community. Although technical details about this harness have not yet been disclosed, its presence indicates DeepSeek's efforts to provide a more complete development ecosystem for users of their models.

From an industry perspective, this release adds competitive pressure on other major AI model providers. DeepSeek's consistent strategy of releasing open weight models with permissive licenses creates an attractive alternative for organizations looking to avoid dependence on paid APIs from providers such as OpenAI or Anthropic. This context is important given the increasingly fragmented global AI landscape. On one side, there is the closed approach relying on APIs and cloud infrastructure; on the other, there is the open source movement supported by communities and companies that want full control over their models. DeepSeek has clearly chosen the latter path.

For developers in Asia and globally, the availability of a model with an MIT license opens opportunities to build applications on a foundation that does not carry licensing costs. This contrasts with models using more restrictive licenses that may limit commercial use or require specific attribution. Although this announcement comes from a credible source and was retweeted by the official Hugging Face account, specific details about the model's architecture, parameter count, or performance benchmarks have not been disclosed in the initial announcement. The research and developer community will likely conduct independent evaluations of the model's capabilities soon. The presence of a real time tracker for DeepSeek V4 at deepseek v4.run indicates high public interest in this release.

Such trackers typically appear when a model release is highly anticipated by a large community, signaling that DeepSeek has built a substantial user base. DeepSeek's move is also relevant in the context of geopolitical AI competition. As a Chinese company, the decision to release a model openly with a permissive license can be seen as an effort to build influence in the global AI ecosystem, regardless of regulatory hurdles that may exist in various jurisdictions. For companies considering adopting this model, security and compliance factors will be important considerations. Although the MIT license removes legal barriers to use, organizations still need to conduct due diligence on the technical and security aspects of the model before integrating it into production systems.

The next development to watch is the release of DeepSeek Harness, which the announcement describes as highly anticipated. If this tool provides comprehensive evaluation and testing capabilities, it would be a valuable addition to the open source AI ecosystem, which currently lacks standardized tools for consistent model benchmarking. With this release, DeepSeek strengthens its position as a key player in the open source AI movement. The combination of open weight models, an MIT license, and plans for supporting tools indicates a long term strategy to build a self sustaining ecosystem attractive to developers worldwide. The AI community now awaits independent evaluations of DeepSeek V4 Pro 0813's capabilities, particularly regarding performance compared to closed commercial models and other open source models.

The results of these evaluations will determine the extent of this release's impact on the adoption of open weight models in the industry.