HUMAIN, a Public Investment Fund company, announced humain-m3, a 428-billion-parameter Arabic-language model, on Sept. 3 at the LEAP conference in Riyadh. The model was built by Chinese AI firm MiniMax and is now available in research preview through HUMAIN Node, the company's developer platform.
The model underwent further pre-training on more than one trillion tokens of Arabic-native content and achieved the highest average score among frontier models across seven public Arabic benchmarks. HUMAIN plans an open-weight release under the MiniMax Community License.
The move reflects a strategic bet by Saudi Arabia on frontier Arabic AI capabilities. Tareq Amin, CEO of HUMAIN, said the model addresses a gap: "Arabic remains underrepresented at the frontier of artificial intelligence, despite hundreds of millions of speakers."
The partnership highlights a fault line in global AI geopolitics. The U.S. and its allies have grown cautious about sovereign AI platforms built on Chinese technology. Yet HUMAIN chose to build on MiniMax's openly available M3 model rather than develop independently—a pragmatic choice that prioritizes speed and capability over technology sovereignty.
The decision carries real tradeoffs. Relying on Chinese model architecture means Saudi Arabia's frontier AI sits on a foundation it does not control. But developing a competitive 400-plus-billion-parameter model from scratch would have taken years and billions more. HUMAIN Node, which will host models from HUMAIN and other providers, attempts to create an ecosystem moat—but only if developers choose to build there.
The open-weight release suggests HUMAIN wants to seed adoption across the Arabic-language developer community. That strategy works only if the model is genuinely competitive with frontier English-language systems. Early benchmark claims are one thing; sustained developer adoption is another.
