MiniMax Photo: VCG

Beyond just application programming interfaces (APIs) and fully developed products, China's artificial intelligence (AI) models are now emerging as foundational tech infrastructure for global developers to build their next-generation AI models, Chinese experts said on Friday, after a Saudi Arabian AI company unveiled a large language model (LLM) based on one from Chinese AI firm MiniMax. 

Humain, a tech firm backed by the Saudi Public Investment Fund, announced on Thursday the release of its Arabic-language model humain-m3, which is based on Chinese AI firm MiniMax's open-source flagship model, MiniMax-M3.

Built on the MiniMax-M3 lineage, humain-m3 is a 428-billion-parameter mixture-of-experts model further pre-trained on more than one trillion tokens of Arabic-native content, according to the company.

MiniMax M3 was released and made open-source in June 2026, with the company positioning it as a flagship model tailored for coding, agentic applications, and native multimodal use cases.

"Unlike in the past, when Chinese LLMs primarily entered overseas markets through APIs or application products, in this collaboration, MiniMax M3 has directly become the foundational model for overseas indigenous model development," Chen Jing, a vice president of the Technology and Strategy Research Institute, told the Global Times on Friday.

According to Chen, Chinese AI firms used to sell products with ready-made capabilities like chatbots and office assistants with limited customization options. This time, a Chinese company is providing the core tech framework and initial model weights, and the Saudi partner gets a controllable, upgradable, and deeply localizable digital infrastructure.

This marks a shift from providing commercial models to the export of technical standards and ecosystem capabilities, Chen noted.

In evaluations across seven public Arabic benchmarks, humain-m3 achieved the highest average score among the frontier models tested, demonstrating strong performance across Arabic language understanding and reasoning, the Saudi company said.

Chen noted that this partnership profoundly demonstrates the global ecosystem value of China's open-source models - particularly in terms of technological accessibility. By providing an open-source foundation, it lowers the barrier for latecomer nations to acquire cutting-edge AI capabilities, enabling them to build sovereign models at a manageable cost, the expert said.

The Saudi company's choice of a Chinese LLM was partly because Chinese open-source models have now joined the global top tier in performance, with their technical capabilities widely recognized, Chen said.

In almost every month of 2026, the largest and most performant open model from a Chinese lab was larger than any model an American lab released, according to a report by Hugging Face on August 14. 

Bloomberg reported in August that Alibaba's open-weight models have accumulated more than 3 billion global downloads in the past six months, eclipsing Meta, Alphabet and other peers to become the world's No.1 AI model. "Qwen's rise suggests that strategy is gaining traction beyond China," the report added.

Open models can be downloaded, customized and used as building blocks for new AI products, making adoption a gauge of which technologies developers are choosing to build on, and that has made download and derivative-model figures one measure of influence in the AI race, according to Bloomberg. 

Chinese AI companies have released a series of major open models recently. On September 1, iFlytek released and open-sourced two general-purpose LLMs, Spark X2.5-4B and Spark X2.5-1.7B, both natively supporting context windows of up to 1 million tokens.

On July 27, Chinese AI company Moonshot AI officially open-sourced the complete model weights of its self-developed LLM Kimi K3. Within just one hour of the release, the model topped Hugging Face's trending chart, setting a new platform record.

China's open-source models have achieved ecosystem autonomy, providing countries in the Global South with a controllable technological foundation. This is driving the global AI governance landscape from a unipolar one to a multipolar, co-existing paradigm, Chen noted.