DeepSeek and MiniMax Get China's Green Light for Global Use of Their AI Models

What happened

On Thursday (22) and Friday (23) this week, the Yuyuan Tantian social media account, linked to Chinese state broadcaster CCTV, posted that DeepSeek and MiniMax received approval from the Chinese government for global use of their artificial intelligence models. The information was reported by Reuters and republished by Economic Times this Friday (23).

One caveat is worth noting, one that the report itself takes care to flag: there is, so far, no official document from the Chinese government confirming the approval. The source is a social media post from an account affiliated with state media, not a statement from the Ministry of Commerce or any regulatory body. Asked for comment by Reuters, DeepSeek and MiniMax did not respond to requests.

In a second post, the following day, the same account said internet users outside China asked for other Chinese open source language models to follow the same path. This suggests the DeepSeek and MiniMax case could (or could not) become a precedent for competitors like Alibaba (Qwen), Zhipu AI (GLM), and 01.AI, though none of this has been officially confirmed.

The October 16 export control package

To understand why this approval became news, it's necessary to go back a week. On October 16, 2026, the Chinese government proposed export controls for 23 AI technologies, including models, agents, and computing clusters. The measure is Beijing's latest response to years of US restrictions on the sale of advanced AI chips to Chinese companies, a move that began in 2022 and has intensified since then.

China's Ministry of Commerce said, at the time, that most AI technologies would remain exportable, with simplified licensing for open source technology exports and for Chinese companies that operate their own data centers abroad. Even so, the proposal covers the deployment, outside China, of models containing restricted technology, which put exactly companies like DeepSeek and MiniMax in question, since they sell API access and distribute weights for international use.

The approval announced by Yuyuan Tantian then works as a signal that these two specific companies passed muster under the new regulatory regime and remain free to operate outside China, at least for now.

Why DeepSeek and MiniMax matter to developers

DeepSeek and MiniMax are not just two more Chinese AI startups: they are two of the few labs outside the United States that distribute open weights for frontier models, competing directly with Meta (Llama) in the open model space and, indirectly, with OpenAI, Google, and Anthropic in the API-hosted model space.

DeepSeek became globally known following the release of DeepSeek-R1, in January 2025, a reasoning model with open weights under a permissive license that became one of the most downloaded on Hugging Face and even sent AI hardware company stocks tumbling in the US market, due to the training and inference cost it claimed. MiniMax follows a similar path, with models like the M1/M2 family, geared toward agentic tasks and long-form reasoning, also distributed with open weights.

For those building products with language models, this kind of lab has become a real lower-cost alternative to the proprietary models of major American players, whether by running locally with tools like Ollama and vLLM, or by calling the API hosted by the Chinese company itself or by third-party providers like OpenRouter and Together AI.

What changes (and what doesn't) in practice

In short: anyone already using DeepSeek or MiniMax weights downloaded from Hugging Face, running on their own infrastructure, is not directly affected by this episode. China's export control targets the operation of Chinese companies abroad, not the use of weights already published under an open license. Even so, there is a difference between the two consumption scenarios for these models:

• Open weights, self-hosted: the regulatory risk falls on DeepSeek and MiniMax, not on whoever has already downloaded and runs the model outside China; China's approval or lack thereof does not revoke a license already granted.
• API hosted by the Chinese company itself: here the dependency is directly on the company remaining authorized to operate the service for international customers, which is exactly what this news is trying to confirm.

Until this week, it wasn't clear whether the October 16 control package would affect the continuity of these hosted services for customers outside China. The approval reported by Yuyuan Tantian indicates that, for DeepSeek and MiniMax, the answer is no, for now, which somewhat reduces uncertainty for those who depend on these APIs in production.

This carries concrete weight for product teams in Brazil evaluating these models as a cost alternative to GPT or Claude: dependency on a provider subject to export controls from a foreign government remains a risk factor worth monitoring, regardless of the model's price or performance.

What remains open

No Chinese authority has publicly confirmed, as of this report's publication, the exact terms of the approval cited by Yuyuan Tantian: it is not known whether a formal license was granted, a specific exemption within the simplified licensing for open source technology, or some other mechanism. There is also no direct confirmation from DeepSeek and MiniMax about the episode; the two companies simply did not respond to Reuters.

It also remains open whether other Chinese open model labs, such as Alibaba's Qwen and Zhipu AI's GLM, will seek (or need to seek) similar approval, and whether this process will be public enough for companies outside China to assess the regulatory risk of each provider before putting one of these models into production.