Abu Dhabi’s Mohamed bin Zayed University of Artificial Intelligence has introduced its largest fully open AI model to date, marking the newest milestone in a years-long campaign to establish the UAE as a genuine contender alongside the United States and China in the global race for artificial intelligence capability.
The release builds directly on a lineage of open-source models developed by MBZUAI’s Institute of Foundation Models, a research arm that has steadily expanded the scale and sophistication of its releases since its first major breakthrough in 2024. That earlier model, K2-65B, was notable for being fully reproducible, meaning every stage of its training process, from data curation to fine-tuning, was published openly rather than kept behind closed doors as is common practice among leading Western AI labs.
A philosophy built on radical transparency
What distinguishes the UAE’s approach from many other national AI efforts is its consistent commitment to what researchers at the university describe as “360-degree openness.” Rather than releasing only a model’s final weights, as many so-called open models do, MBZUAI has repeatedly published full training data, methodology, checkpoints and fine-tuning recipes alongside each release, allowing researchers anywhere in the world to study, reproduce and build upon its work.
That approach carried through to K2 Think, an advanced reasoning model developed in partnership with Abu Dhabi’s AI powerhouse G42, which drew a direct endorsement from UAE President Sheikh Mohamed bin Zayed Al Nahyan when it launched, and which outperformed AI systems many times its size on complex mathematical and logical reasoning benchmarks. A subsequent version, K2 Think V2, released earlier this year as a 70-billion-parameter system, extended that transparency to every stage of the model’s development, from pre-training data through to final evaluation.
Independent analysts take notice
The strategy has not gone unnoticed among independent observers of the global AI landscape. Evaluations by outside AI benchmarking firms have repeatedly ranked MBZUAI’s models at or near the top of openness benchmarks, reflecting the university’s decision to publish its full training methodology rather than the partial disclosures typical of many commercial AI labs.
Policy analysts tracking the Gulf’s AI ambitions have pointed to the sustained scale of investment behind these releases as a key reason for the UAE’s rising profile in the sector. Observers note that the country has built the kind of dedicated compute infrastructure and research talent needed to compete at the frontier of AI development, despite operating on a far smaller population base than either the US or China.
Built at a fraction of typical cost
One recurring theme across MBZUAI’s recent releases has been cost efficiency. Earlier versions of the K2 reasoning system were reported to have been trained using notably fewer high-end AI chips than comparable Western and Chinese systems typically require, a detail researchers have pointed to as evidence that smaller, more resource-efficient teams can still compete at the frontier of AI capability when supported by sustained strategic investment.
That efficiency-focused approach reflects a broader philosophy articulated by MBZUAI’s leadership: that the next wave of meaningful AI progress may come not from simply building ever-larger models, but from smarter, more transparent engineering that other researchers can genuinely learn from and extend.
Part of a broader national AI strategy
The latest model release sits within a much larger UAE push to establish full-stack sovereign AI capability, spanning everything from domestically built data centre infrastructure to specialised language models covering Arabic, Hindi and other regional languages. That strategy has been reinforced by major national initiatives, including large-scale government AI training programmes and a growing network of public-private partnerships between MBZUAI, G42 and international technology partners.
For a university that only began operating a few years ago as the world’s first institution dedicated solely to AI research, the steady cadence of increasingly ambitious open-source releases underscores how rapidly the UAE has moved to translate national ambition into technical output — output that, thanks to its open design, is now being studied and built upon by AI researchers well beyond the country’s borders. Next Article
Comments 1