Abu Dhabi Unveils K2 Think: A Low-Cost AI Model to Rival OpenAI and DeepSeek
A new player has stepped into the global AI race — and it’s coming from the United Arab Emirates.
The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) announced the launch of K2 Think, a cost-efficient reasoning model designed to compete with OpenAI and DeepSeek.
Unlike massive systems such as DeepSeek’s R1 (671 billion parameters) or OpenAI’s undisclosed flagship models, K2 Think runs on just 32 billion parameters. Despite being much smaller, researchers claim it delivers performance on par with leading reasoning models in math, coding, and science benchmarks.
Built on Alibaba’s open-source Qwen 2.5 model and powered by Cerebras AI chips, the system was developed in collaboration with G42, the UAE-based AI firm backed by Microsoft.
Researchers say K2 Think’s strength lies in its approach — combining chain-of-thought fine-tuning and test-time scaling to boost reasoning capabilities. Rather than building another chatbot, MBZUAI aims to position the model for scientific breakthroughs, helping accelerate discoveries in mathematics, coding, and clinical research.
The project underscores the UAE’s growing ambition to emerge as a global AI leader, diversifying its economy beyond oil and competing with established hubs in the U.S. and China.
As Richard Morton of MBZUAI explained, “K2 Think proves that you can do far more with fewer resources — making advanced AI accessible to regions and researchers that lack the scale of Silicon Valley or Beijing.”
