Home Latest Insights | News UAE’s MBZUAI Launches Low-Cost Reasoning Model to Challenge OpenAI and DeepSeek

UAE’s MBZUAI Launches Low-Cost Reasoning Model to Challenge OpenAI and DeepSeek

UAE’s MBZUAI Launches Low-Cost Reasoning Model to Challenge OpenAI and DeepSeek

The United Arab Emirates (UAE) has launched a low-cost reasoning model, joining the global AI arms race.

The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), the UAE’s flagship AI research university, unveiled a new reasoning model on Tuesday that it claims can rival those of OpenAI and Chinese lab DeepSeek—at a fraction of the size and cost.

The model, dubbed K2 Think, boasts just 32 billion parameters, compared with the 671 billion that power DeepSeek’s R1, and was built on Alibaba’s open-source Qwen 2.5 framework. Run and tested on hardware from U.S. chipmaker Cerebras, K2 Think is designed to deliver performance on par with the industry’s largest reasoning systems, despite being significantly smaller. OpenAI has not disclosed parameter counts for its flagship models.

Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird

Tekedia AI in Business Masterclass opens registrations.

Join Tekedia Capital Syndicate and co-invest in great global startups.

Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).

The project was developed in partnership with G42, the UAE’s AI champion backed by Microsoft. Researchers say the model achieved benchmark parity with rivals across math (AIME24, AIME25, HMMT25, OMNI-Math-HARD), coding (LiveCodeBenchv5), and science (GPQA-Diamond) tasks.

How it Works

Hector Liu, director of MBZUAI’s Institute of Foundation Models, attributed K2 Think’s performance to methods such as chain-of-thought supervised fine-tuning—a step-by-step reasoning approach—and test-time scaling, which boosts performance by allocating extra compute during inference.

“What was special about our model is we treat it more like a system than just a model,” Liu said. “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.”

Richard Morton, managing director at the institute, emphasized the scientific focus: “The fact is that the fundamental reasoning of the human brain is the cornerstone of all the thinking process. With this application, instead of taking 1,000 or 2,000 human beings five years to think through a particular question, this vastly condenses that period.”

Why it Matters

Until recently, the AI race has been dominated by the U.S. and China. OpenAI and American Big Tech firms set the early pace with large-scale foundation models, while DeepSeek’s R1 breakthrough earlier this year bolstered China’s standing. The UAE now hopes to insert itself into this competition, using AI not only to diversify beyond oil but also to enhance geopolitical influence.

The country points to G42 as a symbol of this ambition, though its ties to Microsoft have attracted scrutiny in Washington due to G42’s past links with Chinese partners. At the same time, the UAE faces rising regional competition, with neighboring Saudi Arabia investing heavily through Humain, an AI venture backed by its Public Investment Fund.

What’s Next for K2 Think?

If K2 Think fulfills its promise, the UAE could carve out a niche as the hub for cost-efficient, domain-specific AI. MBZUAI’s model could become a tool for emerging markets and smaller economies that lack the infrastructure of Silicon Valley or Beijing by focusing on reasoning for math, science, and clinical research. This would bolster the UAE’s credibility as a scientific power and accelerate the adoption of AI in places historically shut out of the race.

However, there are challenges. While benchmarks are promising, global AI development is driven by firms with enormous capital reserves and infrastructure. If OpenAI and DeepSeek continue to push forward with models that improve efficiency at a massive scale, K2 Think risks being overshadowed. Geopolitical scrutiny of Microsoft’s partnership with G42 could also complicate adoption abroad, particularly in the U.S. and Europe.

K2 Think does not seek to compete head-to-head with ChatGPT or other general-purpose chatbots. Instead, it is targeted at specialized reasoning tasks—where performance, cost, and efficiency matter as much as scale. The release of K2 Think signals a new phase in its ambition for the UAE: positioning itself not only as a consumer of global AI breakthroughs but as a contributor that could shape how scientific reasoning is automated in the years ahead.

No posts to display

Post Comment

Please enter your comment!
Please enter your name here