K2 Think: UAE’s Tiny Yet Powerful AI Model that Challenges the Bigger-is-Better Myth
This 32-billion-parameter model rivals OpenAI and DeepSeek systems, which boast over 200 billion parameters.
Topics
News
- MoneyHash and noon Payments Partner to Boost Local Payments Across the Gulf
- K2 Think: UAE’s Tiny Yet Powerful AI Model that Challenges the Bigger-is-Better Myth
- Apple’s New Products Reinforce its Long-Term Bet on Intelligence and Sustainability
- AI is Going to Create Massive Unemployment, Warns AI Godfather Geoffrey Hinton
- Mistral AI Raises €1.7B in Series C Funding Led by ASML
- 7 in 10 Smart Devices in UAE Homes Are Hackable, Warns Cyber Council

[Image source: Chetan Jha/MITSMR Middle East]
The UAE has unveiled K2 Think, a lean open-source reasoning model developed by MBZUAI and G42. At just 32 billion parameters, it rivals much popular models on benchmarks in math, code, and science. The release positions the UAE as a third pole in an AI race long dominated by the US and China, proving that efficiency, openness, and strategic design can matter as much as scale.
Researchers from the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) announced the launch of K2 Think on Wednesday. This modest-sized model performs on par with the capabilities of OpenAI and DeepSeek, with over 200 billion parameters.
“This is a technical innovation or, in my opinion, a disruption,” Eric Xing, president and lead AI researcher, MBZUAI told WIRED.
The model is being offered for free by Abu Dhabi’s sovereign wealth funds-backed Emirati tech conglomerate G42. The tech giant is running the model on Nvidia’s alternate hardware, a cluster of Cerebras chips.
The model was developed using several thousand GPUs, with the final training run involving 200 to 300 chips. According to Xing, the plan is to integrate K2 Think into a full large language model (LLM) in the coming months.
The model demonstrates an effective combination of several recent technical innovations such as:
- Fine-tuning on long strings of simulated reasoning
- An agentic planning process that breaks problems down in different ways
- Reinforcement learning that trains the model to reach verifiably correct answers
“How to make a smaller model function as well as a more powerful one—that’s a lesson to learn, if other people want to learn from us,” Xing said.
Other nations in the region are also proactively investing in AI infrastructure and research. “Proving that smaller, more resourceful models can rival the largest systems is an achievement that shows how Abu Dhabi is shaping the next wave of global innovation,” said Peng Xiao, CEO of G42, and a MBZUAI board member.
The release aligns with G42’s plan of expanding AI footprint and follows the launch of the 1GW Stargate UAE facility and Microsoft’s $1.5 billion strategic investment in the Emirati tech conglomerate.
PwC estimates that AI could contribute as much as $320 billion to the Middle East’s economy.