Beijing Grants Overseas Export Approvals to DeepSeek and MiniMax AI Models

Beijing has officially cleared artificial intelligence startups DeepSeek and MiniMax to export their large language models to international markets. The two companies successfully completed China's new export registration process on Thursday, granting them the legal right to offer open-source and commercial licenses overseas.[1][7]

The approvals mark the first authorized overseas deployments under a restrictive regulatory framework implemented earlier in October 2026. While the models themselves have not changed technically, the bureaucratic green light allows both firms to compete directly for global enterprise contracts without violating domestic export controls.[2][4]

State broadcaster CCTV announced the milestone late Thursday through its affiliated WeChat account, Yuyuan Tantian. The post confirmed that the two startups navigated the compliance requirements just days after the government finalized its outbound technology rules, signaling a surprisingly fast bureaucratic turnaround.[7]

"Since the policy was announced, two leading companies, DeepSeek and MiniMax, have successfully completed the registration process and obtained open-source and overseas licensing rights," the state-affiliated account stated, highlighting the government's willingness to support domestic champions abroad.[7]

Navigating the new framework

The rapid turnaround suggests Chinese regulators are eager to maintain a foothold in the global open-source ecosystem, even as they tighten internal oversight. DeepSeek and MiniMax represent two of the country's most capable independent AI laboratories, operating outside the immediate umbrellas of tech giants like Tencent or Alibaba.[3][6]

DeepSeek, backed by quantitative trading firm High-Flyer, gained international traction earlier in 2026 with its 236-billion-parameter V2 model. The company aggressively priced its application programming interface to undercut Western competitors, offering comparable performance on coding benchmarks at roughly 10 percent of the cost of established alternatives.[1][3]

MiniMax, founded by former SenseTime executives in 2021, has focused heavily on character-driven conversational agents and multimodal capabilities. The Shanghai-based startup recently launched its abab6.5 model, which processes up to 200,000 tokens of context and powers a suite of popular social applications across domestic and international markets.[2][5]

Until this week, both companies operated in a legal gray area regarding international distribution. The Cyberspace Administration of China requires algorithmic registration for domestic use, but the recent export regulations added a secondary, more stringent layer of scrutiny for any foundational models crossing national borders.[4][6]

The mechanics of export approval

The new export licenses require companies to demonstrate that their training data and model weights do not compromise national security or leak sensitive domestic information. Regulators evaluate the models' safety guardrails, ensuring they align with domestic content restrictions even when deployed on foreign servers.[2][4]

For open-source releases, the regulatory scrutiny is particularly intense. Once model weights are published on platforms like Hugging Face or GitHub, developers lose the ability to recall or patch them, making the initial security audit the only point of regulatory control over the massive 100-gigabyte files.[3][6]

By clearing DeepSeek and MiniMax for open-source distribution, Beijing is signaling confidence in its domestic alignment techniques and safety filters. The approval allows these models to be freely downloaded and modified by international developers, embedding Chinese architecture deeply into global software supply chains.[1][5]

The registration process also mandates that companies maintain detailed logs of their overseas commercial deployments. Regulators want visibility into which foreign enterprises are utilizing Chinese foundational models, ensuring that the technology is not being repurposed for military or intelligence applications by adversarial nations.[2][6]

The timing of the announcement coincides directly with escalating technology trade tensions. Just hours before the Chinese approvals were publicized, the United States Treasury Department proposed finalized regulations to curb American investments in Chinese artificial intelligence and semiconductor sectors.[7]

Competing in a fractured market

The reciprocal restrictions threaten to bifurcate the global AI landscape, forcing international enterprises to choose between distinct technology stacks. However, the open-source nature of DeepSeek's releases complicates this division, as developers worldwide routinely blend architectures to build highly customized hybrid systems.[2][6]

Industry analysts note that while the export approvals are a significant bureaucratic hurdle cleared, the commercial reality remains challenging. Western enterprises may hesitate to integrate Chinese-origin models into sensitive workflows, citing data privacy concerns, opaque training methodologies, and the ever-present risk of future sanctions.[4][5]

To mitigate these concerns, both startups have emphasized their models' mathematical and coding proficiencies rather than their political or cultural reasoning. Code generation relies on objective logic, making it inherently less susceptible to the ideological biases that regulators on both sides of the Pacific heavily scrutinize.[1][3]

DeepSeek's pricing strategy also serves as a powerful wedge into cost-sensitive markets where geopolitical concerns take a back seat to unit economics. By offering API access at roughly $0.14 per million input tokens, the company has attracted a substantial developer base in emerging economies across Southeast Asia and the Middle East.[3][4]

MiniMax is similarly leveraging its unique strengths, marketing its advanced voice and character-generation capabilities to international gaming and entertainment studios. The startup's ability to generate highly emotive, context-aware dialogue offers a specialized alternative to the more generalized enterprise models dominating the Western market.[2][5]

Setting a precedent for the industry

The successful registration of these two models establishes a clear compliance pathway for other Chinese AI developers waiting in the wings. Startups like Moonshot AI and Zhipu AI are expected to follow suit, submitting their own frontier models for export review in the coming weeks.[2][6]

For now, DeepSeek and MiniMax hold a distinct first-mover advantage in the approved export market. Their ability to legally license their technology abroad allows them to sign binding commercial contracts, moving beyond the experimental phase into full-scale enterprise deployment without regulatory overhang.[1][5]

The true test of this regulatory milestone will be actual enterprise adoption. While Beijing has opened the door for these models to leave the country, the startups must now convince a skeptical global market that their technology is both secure and superior to domestic alternatives.[4][6]