Analysis
HUMAIN, the Public Investment Fund-backed AI company chaired by Saudi Arabia's Crown Prince, unveiled humain-m3 at LEAP Riyadh on Thursday -- a 428-billion-parameter mixture-of-experts Arabic language model built on top of Chinese lab MiniMax's open-weight M3 architecture rather than developed independently, according to Bloomberg and PR Newswire. HUMAIN trained the base model on more than a trillion Arabic tokens, and its own evaluation puts the previewed checkpoint at an average of 89.37% across seven public Arabic benchmarks -- the highest among frontier models tested, per HUMAIN's own testing.
HUMAIN launched in May 2025 under CEO Tareq Amin, a Jordanian-American technologist and former Aramco Digital chief, with a mandate to build a full-stack Saudi AI company spanning infrastructure, cloud, models, and applications. Its closest regional rival is Abu Dhabi's G42, which operates under a similar sovereign-wealth structure, while the Technology Innovation Institute's Falcon-H1 currently tops the open Arabic-model leaderboard independent of HUMAIN's own claims, ahead of Microsoft-backed ALLAM and Qatar's Fanar.
Speed over independence
The choice to build on MiniMax rather than Meta's Llama, a US lab's weights, or a homegrown architecture is the real story: it signals Gulf sovereign-AI projects increasingly treat Chinese open-weight models as the fastest, cheapest foundation layer available, ahead of anything from the US labs that dominate the closed-model conversation. That's a meaningful shift given the geopolitics involved -- Saudi Arabia has spent years balancing US and Chinese technology partnerships, and a national AI platform built on Chinese weights, even open ones, complicates the narrative that Gulf AI investment flows exclusively toward US compute and US models.
The unverified part of the claim matters too: HUMAIN's 89.37% average score comes from its own internal evaluation of a research-preview checkpoint, not an independent third party, and Tech Times' own reporting flagged the scores as unverified pending outside testing. Sovereign AI announcements have a track record of front-loading benchmark claims that don't hold up once external researchers get access -- worth remembering before treating humain-m3 as a settled leader over Falcon-H1 or ALLAM.
