Pramana, MiniMax get China’s global AI approval: report

China’s reported approval means Pramana and MiniMax can use or offer their AI models globally under the state’s approval framework. It does not mean every Chinese AI technology is freely exportable. Beijing is separately proposing controls on technologies it considers important to national security.

The reported approval was announced by Yuyuan Tantian, a CCTV-affiliated social media account, on October 22. It named Pramana and MiniMax, a China-based AI company listed in Hong Kong. The announcement followed China’s October 16 proposal to control exports of 23 AI technologies.

The situation remains subject to government rules. China’s commerce ministry said most AI technologies would remain exportable and that licensing could be simplified for open-source exports and overseas computing centers. Pramana and MiniMax had not immediately commented, so the companies’ detailed approval conditions are not public.

The two companies named in the reported approval are Pramana and MiniMax. Pramana is identified as a Chinese AI developer. MiniMax is described as a China-based AI company listed in Hong Kong. Both are connected to the development of AI models.

The article places them within China’s broader open-source large language model community. It says overseas internet users urged other Chinese open-source large language model developers to seek similar approval. This indicates that the approval concerns models intended for use beyond China, rather than only domestic AI products.

The article does not describe the specific capabilities, training data, or model sizes of Pramana’s and MiniMax’s systems. It also does not say whether every model from either company received approval. The companies did not immediately respond to requests for comment, so the exact scope of the reported authorization remains unclear.

China proposed export controls on 23 AI technologies on October 16. These controls are part of a broader effort to regulate the overseas transfer of tools that Beijing considers important to national security.

The proposal does not mean that all AI technology would be blocked from leaving China. The commerce ministry said most AI technologies would remain exportable. However, technologies covered by the proposed restrictions could require licensing before export or overseas deployment.

The controls also cover overseas deployment of AI models containing restricted technology. That detail is important because a company might face rules even when it is operating a model outside China. Against this background, the reported approval for Pramana and MiniMax signals that global use of some Chinese AI models may be permitted, while other technologies remain subject to tighter review.

Beijing is trying to limit overseas transfers because it considers some AI tools vital to national security. The policy aims to keep control over technologies that could have strategic importance when shared with foreign users, companies, or governments.

China’s October 16 proposal covers 23 AI technologies. It also covers overseas deployment of AI models containing restricted technology. This means the rules may apply not only when technology is exported as software or equipment, but also when a model using restricted technology is operated abroad.

The policy does not seek to stop every AI export. China’s commerce ministry said most AI technologies would remain exportable. It also said licensing for open-source technology exports and for Chinese companies operating their own overseas computing centers would be simplified. The key distinction is between ordinary or supported exports and technologies Beijing classifies as sensitive.

If overseas exports or operations require approval, Chinese AI companies may need to wait for licenses before serving foreign users. That could slow launches, complicate partnerships, and make international planning less predictable. These are possible business effects of a permission-based system, not outcomes confirmed in the article.

The proposed controls cover exports of 23 AI technologies and overseas deployment of models containing restricted technology. A company would therefore need to determine whether its model includes controlled technology and then follow the applicable licensing process. The reported approval for Pramana and MiniMax shows that global use may still be authorized.

The effect will depend on how China applies the rules. The commerce ministry said most AI technologies would remain exportable and promised simpler licensing for open-source exports and company-run overseas computing centers. Tighter treatment would likely focus on models or tools classified as sensitive, while other international activity could continue more easily.

Simplified licensing means eligible companies may face a more streamlined approval process rather than the strictest controls. It could help Chinese firms export open-source AI technology or run their own overseas computing centers with fewer administrative barriers, subject to the rules.

China’s commerce ministry specifically mentioned simplified licensing for open-source technology exports and for Chinese companies operating their own overseas computing centers. By contrast, the proposed controls cover 23 AI technologies and overseas deployment of AI models containing restricted technology. Those activities would remain under closer review.

The difference is therefore based on the technology and the activity involved. Open-source exports and company-operated computing centers may receive an easier licensing route. Models that contain restricted technology may require more scrutiny or face limits. The article does not provide the detailed criteria, timelines, or approval conditions for either category.

A large language model, or LLM, is an AI system trained to process and generate human language. It learns patterns from very large collections of text and uses those patterns to predict useful sequences, such as answers, summaries, or translations. This explanation uses established general knowledge because the article names LLMs but does not define them.

Training data influences what information, languages, styles, and biases a model can reflect. Computing power affects how much data and how many model parameters can be processed during training, as well as how quickly the system can respond. More computing does not automatically guarantee accuracy, since data quality and software choices also matter.

Software determines the model’s architecture, training methods, safety controls, and tools. Together, these elements shape what the model can do and how reliably it does it. In the article’s context, such models matter because China’s proposed controls may cover models containing restricted technology when they are deployed overseas.