HUMAIN, the new national AI company formed by Saudi Arabia’s Public Investment Fund, will release an Arabic-centric foundation model called ALLaM 34B by the end of August, CEO Tareq Amin told Middle East AI News. The model has been trained and fine-tuned for Arabic language, regional dialects, and Arab cultural context, with a focus on Saudi Arabia. An entire HUMAIN Chat application suite has been built around the model and will be available to the public.
Amin said the model is a clean-slate foundation system rather than an iteration of earlier SDAIA work. HUMAIN has taken over development and commercialisation of ALLaM from SDAIA following the merger that combined SCAI, SDAIA’s model team, and talent from Aramco Digital and Tonomus. According to an interview with Asharq Al Awsat, a 40-strong PhD research team trained the new model on proprietary in-Kingdom datasets.
HUMAIN’s remit spans next-generation data centres, advanced AI infrastructure, and Arabic AI models. In May the company introduced HUMAIN OS, a platform that offers a unified Arabic and English conversational interface for enterprise applications. HUMAIN’s partner Groq this month made OpenAI’s new open-weight gpt-oss models available in Saudi Arabia on day zero via GroqCloud servers in Dammam, and HUMAIN Chat is expected to run on GroqCloud as well.
SDAIA released the first ALLaM models in 2023 and 2024, including ALLaM 13B on IBM’s watsonx in May 2024 and later via Microsoft Azure in September 2024. The new ALLaM 34B is positioned as a larger, general-purpose foundation model intended for sovereign, enterprise, and research use.
The regional backdrop is increasingly competitive. Since 2023, sovereign programmes in the Gulf have rushed to produce Arabic large language models, and many local developers have also trained or fine-tuned their own systems to meet specific product needs.
Public AI cloud services from regional players have yet to reach the breadth of offerings from global providers, which has kept deployment options fragmented. HUMAIN’s launch aims to close that gap with a model and app stack tailored for Arabic users and in-region infrastructure.
