- MBZUAI announces global multi-site AI institute, with a new center in California alongside existing ones in Abu Dhabi and Paris
- Unveils PAN, a groundbreaking “world model” capable of simulating real-world scenarios with multi-modal inputs
-
Ongoing development of advanced AI models “K2-65B” and “Jais”, reinforcing transparency and open-source AI development
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has announced the launch of the Institute of Foundational Models (IFM) a major new initiative featuring a global network of research centers. The announcement was made during an event at the Computer History Museum in Silicon Valley, unveiling a new research center in Sunnyvale, California, to join existing MBZUAI research hubs in Paris and Abu Dhabi.
The new institute aims to accelerate the global development of foundational AI models and deepen engagement with the thriving AI ecosystem in California, which includes top researchers, startups, and leading technology companies.
Professor Eric Xing, President of MBZUAI, stated:
“The launch of IFM is a significant step toward global collaboration on cutting-edge AI research. Our expansion into Silicon Valley places us at the heart of one of the world’s most active AI ecosystems, opening the door for meaningful partnerships and real-world applications.”
The event featured demonstrations of “PAN” (Predictive Agent for the Natural world), a world model capable of simulating complex real-world scenarios using multi-modal data—text, video, spatial information, and physical dynamics. Unlike traditional models focused on single output formats, PAN generates comprehensive state predictions, enabling advanced reasoning, planning, and decision-making. Applications range from autonomous driving to robotics.
A key innovation, the PAN-Agent, showcases PAN’s ability to solve multi-modal reasoning tasks like math and code generation within dynamic simulation environments.
The institute also announced two flagship models under development:
-
K2-65B: A large language model focused on advanced logical reasoning, math solving, and code generation, optimized for efficient performance with lower compute costs
-
Jais: One of the most advanced Arabic LLMs globally, supporting Modern Standard Arabic, dialects, and languages such as Hindi—while preserving cultural context and expanding multilingual capabilities
MBZUAI emphasized its commitment to transparency and open development, offering full research documentation, datasets, and model checkpoints via its initiative LLM360, making it a global benchmark in responsible AI research.
The institute combines the agility of startup innovation with the research rigor of global academic institutions. Dedicated teams focus on model design, training methods, evaluation frameworks, and AI safety protocols, aiming to translate scientific breakthroughs into real-world impact.
العربية (Arabic) To read the article in Arabic, click here
