The United Arab Emirates (UAE) has unveiled a groundbreaking AI model that showcases enhanced reasoning capabilities, rivaling the top technologies from both the United States and China. This progress is a significant indicator that the UAE’s considerable investments in artificial intelligence are beginning too yield impressive results.
The new model, named K2 Think, has been developed by researchers at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), located in Abu Dhabi, the capital of the UAE. This model represents one of the first “sovereign” AI systems designed with advanced reasoning capabilities and is being made available for free by G42, an Emirati technology company supported by Abu dhabi’s sovereign wealth funds. G42 operates this model on a cluster of Cerebras chips as an alternative to Nvidia’s hardware.
K2 Think stands as one of the UAE’s key contributions to the global race for technological supremacy in artificial intelligence—a field anticipated to have significant economic and geopolitical ramifications. While the US and China are recognized as leading players in this arena, numerous smaller nations with considerable financial resources are also striving to develop their own “sovereign” AI systems.
This particular model features 32 billion parameters, making it relatively compact compared to its counterparts. Unlike conventional language models that generate responses quickly based on data synthesis, K2 Think specializes in reasoning tasks through a simulated deliberative process. According to its developers, it performs comparably to reasoning models from OpenAI and DeepMind that boast over 200 billion parameters.
“This represents not just technical innovation but perhaps a disruption,” stated Eric Xing, president of MBZUAI and lead researcher on this project during an interview with WIRED prior to today’s declaration.
Xing emphasized that K2 Think exemplifies an effective blend of several innovative techniques including fine-tuning over extensive sequences for simulated reasoning; an agentic planning approach that dissects problems from various angles; and reinforcement learning aimed at achieving verifiably accurate answers. Additional innovations enable efficient operation on Cerebras chips.
“The ability to create a smaller yet more powerful model is something others can learn from us,” Xing remarked.
He further noted that K2 Think was developed using thousands of gpus (though he did not disclose specific figures), with final training involving between 200-300 chips. The goal is for K2 Think eventually to be integrated into a thorough large language model (LLM) within upcoming months. MBZUAI has open-sourced this model and published a technical report detailing how various innovations were combined during its development.
Other nations within the Middle East—including Saudi Arabia—are also heavily investing in AI infrastructure and research initiatives. In May, President Donald Trump visited the region announcing several collaborative projects involving US tech companies focused on AI advancements.
The UAE government has committed billions towards establishing itself as a pivotal research hub for technology innovation.The country has already produced cutting-edge AI research while setting up operations in Silicon Valley itself. Additionally, it has reduced its reliance on China, seeking access instead to american silicon necessary for developing advanced technologies.
Peng Xiao, CEO of G42 and board member at MBZUAI commented: “By demonstrating that smaller yet more efficient models can compete with truly large systems, this achievement illustrates how Abu Dhabi is influencing future global innovation.”