The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi has unveiled a new low-cost reasoning model called K2 Think, seeking to rival AI models from DeepSeek and OpenAI.
This development comes after DeepSeek claimed in January that it was nearing OpenAI’s capabilities with a significantly lower budget and energy consumption. K2 Think stands out by using fewer parameters than its competitors while achieving comparable performance in reasoning tasks.
MBZUAI characterizes K2 Think as “a new class of reasoning model” that employs long chain-of-thought supervised fine-tuning and reinforcement learning with verifiable rewards to improve logical depth and accuracy. According to Hector Liu, director of MBZUAI’s institute of foundation models, K2 Think is treated as a dynamic system, continually enhanced through deployment and iterative adjustments.
As per MBZUAI, K2 Think is among the fastest and most efficient reasoning systems available, capable of processing around 2,000 tokens per second, equivalent to approximately 1,500 words. The model is built upon Alibaba’s Qwen 2.5 large language model and operates on hardware provided by AI chipmaker Cerebras.
Similar to DeepSeek’s R1 model, K2 Think is open source, making its training data and weights publicly accessible. This transparency, as stated by the university, will enable the global research community to examine, replicate, and expand upon the model’s reasoning capabilities.
MBZUAI notes that K2 Think represents a significant milestone for AI development in the UAE, exemplifying how open innovation and public-private partnerships can establish Abu Dhabi as a prominent global AI leader. The university believes that ingenuity and collaboration are crucial factors in shaping the future of AI reasoning.
The introduction of K2 Think could have a notable impact on the global AI landscape, demonstrating the UAE’s potential to compete with leading nations like the U.S. and China through strategic innovation and partnerships.
