DeepSeek and MiniMax Complete AI Export Registration and Begin International Licensing
Read more
Pandaily
pandaily.com

DeepSeek and MiniMax Complete AI Export Registration and Begin International Licensing

Two prominent Chinese artificial intelligence model developers, DeepSeek and MiniMax, have successfully completed the registration process introduced as part of the proposed update to the country's export technology catalog. Both companies have continued their operations by releasing open-source models and initiating international licensing.

This information was revealed on the evening of October 22nd via the Yuyuan Tantian account belonging to the state broadcaster China Media Group. These developments occur against the backdrop of a project to revise the Catalog of Prohibited or Restricted Export Technologies, which the Ministry of Commerce and the Ministry of Science and Technology put out for public comment on October 16th.

The draft changes included 23 items related to AI, covering large models, AI agents, computing clusters, and information processing. The catalog itself has already drawn attention, and this new information confirms that two leading model developers have already applied this process in practice.

It is important to note that registration differs from obtaining a restrictive license, and this catalog is currently a consultative draft, not a final list. According to the Ministry of Commerce, technologies that have already become publicly available through open-source releases or academic publications are proposed to be excluded from export regulation.

Automatic licensing is provided for cases where Chinese companies build and operate overseas data centers, as well as for joint research utilizing foreign resources; the ministry describes this process as being close to export after an application is submitted.

The Ministry has also defined the boundaries of how models reach users abroad. Providing a restricted model via API calls, applications, or web interfaces, as well as in the form of SaaS products or light local deployment on a device, is not considered the export of large model technology. However, local deployment of a restricted model abroad falls under export control, although placement on the company's own overseas cloud infrastructure may utilize automatic licensing.

The Ministry emphasized that the vast majority of AI technologies will remain freely exportable.

The practical steps taken by the companies are giving concrete shape to the process. As reported by Huxiu, DeepSeek opened its framework for large-scale inference, DeepInfer, and the accompanying deepseek-recipe toolkit on October 20th. MiniMax, meanwhile, set up an official application pathway for foreign regions previously excluded from its model licenses, including the United States, the European Union, the United Kingdom, and South Korea.

Under this approach, model weights remain open, but organizations in these regions must undergo compliance checks and obtain a license before deploying the models locally. For developers, this means that open weights, open source code, and API access will likely remain as they are, while local deployment of the most sensitive technologies abroad will require an application procedure. The discussion period for the catalog draft is still ongoing, and the final scope of regulation may change before it comes into effect.

Similar stories

DeepSeek Founder Liang Wenfeng Meets with Embodied AI Company Founders, Considering Entry into Robotics
Read more
pandaily.com

DeepSeek Founder Liang Wenfeng Meets with Embodied AI Company Founders, Considering Entry into Robotics

DeepSeek founder Liang Wenfeng has recently been paying close attention to embodied artificial intelligence and held discussions with several founders in this field, according to Phoenix Tech on October 12. The discussions covered topics such as robotics, large models, and the interaction of AI systems with the physical world.

According to sources familiar with the situation, Liang is studying the possibility of further advancing DeepSeek into the field of embodied AI. Two possible directions are currently being discussed: either creating its own team for research and development or entering this market through investments in already operating companies. No decisions have been announced at this time.

One of the interviewees from Phoenix Tech noted that Liang does not have extensive experience in this area and is intensively studying it. At the time of publication, Sina Tech had not received a response from DeepSeek to requests for comment.

Part of the interest in this direction is related to recent achievements in other fields. According to a report, several demonstration tests showed that OpenAI's GPT-6 achieves task completion results in embodied scenarios, which surprised the industry, and this is one of the reasons influencing DeepSeek's considerations.

Another industry representative quoted in the article suggested that such a move would not be unexpected. He mentioned that Zhipu and Moonshot AI, the company behind Kimi, already have embodied AI developments in reserve, and that embodied intelligence has become a new focus area for AI companies worldwide.

For DeepSeek, entering robotics would signify a significant expansion of its operations. Until now, the company has focused primarily on language models and related software. In recent weeks, DeepSeek released DeepSeek-V4.1-Flash with native multimodal input support, opened its agent runtime Harness, and presented versions of its core inference libraries for Ascend. Liang has also publicly stated that the company's future models should be trained on domestic chips.

Embodied AI requires different capabilities: data collected by physical robots, simulation, hardware partners, and control models capable of real-time operation. Regardless of whether DeepSeek decides to develop everything independently or work with partners, it is reported that the company is exploring how its models can go beyond the screen and transition into machines. For now, plans remain in the consideration stage.

Popular