The UAE has taken a step into the AI spotlight with K2 Think, a low-cost reasoning model developed at MBZUAI in Abu Dhabi.
The system is designed to rival ChatGPT and China’s DeepSeek, with performance on par with much larger models.
MBZUAI’s K2 Think Sets Benchmark at 1,500 Words per Second
“What was special about our model is we treat it more like a system than just a model,” Hector Liu, director of MBZUAI’s Institute of Foundation Models, told CNBC.
The university said K2 Think is capable of processing roughly 1,500 words per second and is open-source, making its learning process fully transparent to researchers worldwide.
Introducing a New Class of Reasoning Model
K2 Think is described by MBZUAI as “a new class of reasoning model.” It employs long chain-of-thought supervised fine-tuning to enhance logical depth and reinforcement learning with verifiable rewards to sharpen accuracy on difficult problems.
Built on Alibaba’s Qwen 2.5 large language model and run on Cerebras hardware, K2 Think is smaller in terms of parameters compared to its competitors, including DeepSeek.
The university says its reasoning capabilities are on par with larger models.
“This new level of transparency ensures that every step of how the model learns to reason can be studied, reproduced, and extended by the global research community,” MBZUAI said in a press release.
Implications for UAE’s AI Ambitions
The launch of K2 Think positions Abu Dhabi as a potential hub for artificial intelligence innovation. MBZUAI emphasises that the project demonstrates how collaboration between government, universities, and private companies can produce world-class AI solutions.
“K2 Think is a defining moment for AI in the UAE. It reflects how open innovation and close public–private partnerships can position Abu Dhabi as a global leader in AI, demonstrating that the future of reasoning will be shaped not only by size, but by ingenuity and collaboration,” the university said.
Challenging the Global AI Paradigm
K2 Think challenges the notion that only massive models dominate AI development, opening opportunities for low-cost, high-efficiency reasoning systems. With its open-source design, it invites the international research community to study, reproduce, and extend the model.
Technical Capabilities and Infrastructure
- Processes 2,000 tokens per second (~1,500 words)
- Uses long chain-of-thought supervised fine-tuning and reinforcement learning
- Built on Alibaba’s Qwen 2.5 LLM
- Run on Cerebras hardware
- Open-source: training data and weights publicly available
UAE’s Forward-Looking AI Strategy
The Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi has introduced K2 Think, a cost-efficient AI model aimed at competing with global leaders. The university said it is “one of the fastest and most efficient reasoning systems in existence.”
Observers say the launch highlights the UAE’s growing influence in artificial intelligence beyond the traditional US–China axis. Abu Dhabi is using innovation and collaboration to be a valuable player in the international AI community.