News
Saudi Arabia's Flagship AI Model humain-m3 Turns Out to Be Chinese MiniMax-M3
Humain, a company backed by Saudi Arabia's PIF sovereign wealth fund, unveiled humain-m3 as the foundation of a national AI push, but its own press release admits the model was built on Chinese MiniMax-M3 architecture.
Contents
Saudi company Humain, backed by the government's Public Investment Fund, unveiled its humain-m3 language model on September 3, 2026 at the LEAP conference in Riyadh. The company presents it as the foundation of a national artificial intelligence platform for the Arabic-speaking world, but its own press release acknowledged that the model was built on an architecture from the Chinese company MiniMax.
Humain's press release describes humain-m3 as a model commissioned by Humain and delivered by MiniMax. That phrase, from the first paragraph of the official announcement, settles any doubt about the technology's origin, even though earlier reports suggested the company had presented the project as fully homegrown until observers noticed that its parameter count matched MiniMax-M3 exactly.
What Humain showed
The model is currently available to the public as a research preview through the HUMAIN Node platform, the company's own interface offering a single API key and a single account for multiple models, with a choice of global hosting, hosting within the Kingdom, or fully sovereign hosting. humain-m3 supports three reasoning modes, tool use, and long, multi-step agentic tasks in both Arabic and English.
According to figures released by Humain, the model achieved the highest average score among the systems tested across all seven Arabic-language benchmarks it was measured on. The company places that result at the center of its narrative about building an edge in languages that remain underrepresented at the frontier of artificial intelligence development.
Chinese roots instead of a clean slate
MiniMax-M3 is an open architecture from the Chinese company MiniMax, released as open source in June 2026 and promoted by its creator as a flagship model for coding tasks, agentic work, and native multimodality. Rather than building its own language model from scratch, Humain took the open MiniMax-M3 base and ran additional training on more than a trillion tokens of Arabic content.
That marks a clear shift away from years of investment in independent foundation-model development and toward faster adaptation of an existing architecture. To market observers, it signals that even countries with ambitions of building fully sovereign artificial intelligence are now turning to Chinese open weights as a starting point instead of spending years developing their own architecture.
In the past, Chinese large language models entered foreign markets mainly through APIs or application products. In this collaboration, MiniMax M3 has become directly the base model for building homegrown models abroad - Chen Jing, analyst quoted by Global Times
That shift matters beyond a single model. Instead of exporting finished products or API access, Chinese companies are starting to supply the underlying infrastructure itself, on top of which other countries build their own, formally independent artificial intelligence systems.
Saudi infrastructure, Chinese model foundation
Humain was founded in May 2025 at the initiative of Crown Prince Mohammed bin Salman and is part of Saudi Arabia's Vision 2030 strategy, aimed at weaning the country's economy off oil. The company is led by Tareq Amin, and its operations rely on partnerships with leading Western hardware suppliers, including Nvidia, AMD, AWS, and Mistral.
Humain's first computing cluster uses 18,000 Nvidia GB300 chips, equivalent to 72,000 Blackwell Ultra accelerators with 120 megawatts of capacity. A contract with AWS worth more than $5 billion covers the construction of a dedicated AI Zone, and the AMD portfolio includes Instinct, Epyc, Pensando, and Ryzen AI processors. This Western hardware infrastructure runs in parallel with the Chinese software foundation of the company's latest language model.
Arabic is spoken by hundreds of millions of people, and yet it remains starkly underrepresented at the frontier of artificial intelligence - Tareq Amin, CEO of Humain
What this means for other countries
Humain's example points to an alternative path for countries building their own AI ambitions, including Poland with projects such as Bielik and its push for an EU AI gigafactory. Instead of months spent training a model from scratch, further training an open, ready-made architecture on local-language data allows competitive results in a given language or domain to be reached faster and more cheaply.
At the same time, the humain-m3 case shows how fluid the concept of technological sovereignty in artificial intelligence has become. A country can control a model's infrastructure, data, and deployment while still basing its core on an architecture developed by a geopolitical rival of its main hardware suppliers.
