The United Arab Emirates (UAE) on Tuesday, September 9, launched a low-cost AI reasoning model as an alternative to similar offerings from the likes of OpenAI and DeepSeek.
K2 Think was developed by the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), an AI-focused research university established by the UAE, in partnership with domestic AI firm G42 backed by Microsoft.
It has been built on top of Chinese tech giant Alibaba’s open-source Qwen 2.5 model and runs on infrastructure provided by AI chipmaker Cerebas, according to a report by CNBC. The newly launched AI model has a total of 32 billion parameters, making it much smaller than DeepSeek’s R1 which has 671 billion parameters. The parameter count of a large language model (LLM) reflects its capacity to learn and generate complex responses.
With the launch of K2 Think, the UAE is looking to position itself as a challenger in the global race for AI domination that has, so far, been framed as a contest largely between the US and China. It comes as countries worldwide, including India, ramp up efforts to achieve AI sovereignty.
“What was special about our model is we treated it more like a system than just a model. So, unlike a regular open-source model where we can just release the model, we actually deploy the model and see how we can improve the model over time,” Hector Liu, director of MBZUAI’s Institute of Foundation Models, was quoted as saying by CNBC.
K2 Think is reportedly designed to enable scientific breakthroughs. “The fact is that the fundamental reasoning of the human brain is the cornerstone of all the thinking process. With this particular application, instead of taking 1,000, 2,000 human beings five years to think through a particular question, or go through a particular set of clinical trials or something like that, this vastly condenses that period,” Richard Morton, managing director for MBZUAI’s Institute of Foundation Models, was quoted as saying.
Model specs, performance
Although its source code and core technology seem to be based on Alibaba’s Qwen 2.5 open-source AI model, K2 Think has been further optimised to achieve higher levels of performance with limited computing resources. This has been done through a number of techniques such as chain-of-thought (CoT) supervised fine-tuning and test-time scaling, as per the report.
Story continues below this ad
The model was evaluated on several math, coding, and science-related benchmarks such as AIME24, AIME25, HMMT25, OMNI-Math-HARD, LiveCodeBenchv5, and GPQA-Diamond. In terms of performance, the researchers behind K2 Think claimed that the model matched cutting-edge AI reasoning models from OpenAI and DeepSeek despite being a fraction of their sizes.