The United Arab Emirates has been trying to position itself as a global leader in AI, aiming to boost its geopolitical influence and diversify its economy beyond crude oil dependence.
The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), a research-focused AI university established by the UAE, announced on Tuesday the launch of a new low-cost reasoning model designed to rival OpenAI and DeepSeek, CNBC reports.
This comes after China’s DeepSeek shocked the world earlier this year with the release of its R1 reasoning model, which it claimed could outperform OpenAI—while being trained at a fraction of the cost.
With just 32 billion parameters, MBZUAI’s model, called K2 Think, is much smaller than competing systems from OpenAI and DeepSeek. It was built on Alibaba’s open-source Qwen 2.5 model and is run and tested on hardware supplied by AI chipmaker Cerebras.
For context, DeepSeek’s R1 has a total of 671 billion parameters—essentially variables that an AI language model learns in order to understand and generate language. OpenAI does not disclose the number of parameters in its AI models.
K2 Think was developed in partnership with G42, the well-known UAE AI company backed by U.S. tech giant Microsoft. Researchers behind the model claim it delivers performance comparable to the flagship reasoning models of OpenAI and DeepSeek despite its smaller size. They cited benchmarks such as AIME24, AIME25, HMMT25, and OMNI-Math-HARD (math), the LiveCodeBenchv5 coding benchmark, and the GPQA-Diamond scientific benchmark.
Hector Liu, director of MBZUAI’s Institute of Foundation Models, told CNBC that the K2 Think team reached such high performance levels using a series of methods.
These include supervised fine-tuning of long chain-of-thought (CoT) reasoning—a step-by-step method—as well as so-called test-time scaling, a technique that improves performance by allocating additional computing resources during inference (applying learned knowledge to data the model has never seen before).
“What was special about our model is that we treat it more like a system than just a model,” Liu told CNBC. “So, unlike a typical open-source model that we can simply release, we deploy the model and then see how we can improve it over time.”
“If you ask me which individual step is the most important, it’s very hard for me to answer. It’s more of a systems method, where all these methods combined led to the final result,” he added.
Two countries stand out as pioneers in the global AI race: the U.S. and China.
American tech giants and startups like OpenAI led the initial push with so-called foundation models, designed to handle a wide range of tasks based on vast training datasets. However, DeepSeek’s R1 breakthrough earlier this year cemented China’s position as a formidable AI player.
More recently, the UAE has sought to position itself as a global leader in AI, aiming to expand its geopolitical influence and diversify its economy beyond oil.
The country can point to its AI firm G42 as an example of how it is gaining ground. However, it faces stiff competition from neighboring Saudi Arabia, which aims to build full-scale AI capabilities through Humain, a company launched in May by the Public Investment Fund.
Beyond that, geopolitical complexities also cloud the UAE’s AI ambitions. Microsoft’s investment and partnership with G42 last year drew significant attention in the U.S. regarding the company’s ties to China.
Overall, the UAE’s AI industry still has a long way to go to catch up with its counterparts in the U.S. and China. OpenAI and major tech companies have enjoyed a considerable head start with their AI models, while Beijing has long considered AI a strategic priority.
Although K2 Think demonstrates performance comparable to OpenAI, its developers say the goal is not to build a chatbot like ChatGPT. Richard Morton, managing director of MBZUAI’s Institute of Foundation Models, explained that the model is intended for use in specific fields such as mathematics and science.
“The truth is that fundamental reasoning in the human brain is the cornerstone of the entire thinking process,” Morton told CNBC.
“With this particular application, instead of needing 1,000 or 2,000 people to think about a single question for five years or go through a series of clinical trials or something similar, that time period is condensed significantly.”
It could also broaden access to advanced AI technologies in regions that lack the capital and infrastructure available to U.S. firms.
“What we’re finding is that you can do much more with much less,” Morton said.
Enterprise AI that delivers: secure, sovereign-by-design, predictable, at scale. From POC to production with DenseMAX appliances, Digital Automation, and Surrogate Agents.
