G42 & UAE's AI University Unveil Open-Source AI Reasoning Model that Rival OpenAI, Google
The model delivers performance on par with the largest reasoning models while using just 32 billion parameters.s AI

ABU DHABI, September 9, 2025 — The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) and G42 have announced K2 Think, an open-source system designed to push the boundaries of AI reasoning. Developed by MBZUAI’s Institute of Foundation Models, the new system delivers performance on par with the largest reasoning models while using just 32 billion parameters—20 times smaller than some flagship models.
K2 Think is built on six pillars of innovation, including long chain-of-thought supervised fine-tuning, reinforcement learning with verifiable rewards, agentic planning, and test-time scaling techniques.
The system is optimised for Cerebras’ wafer-scale inference platform, with speculative decoding allowing throughput of up to 2,000 tokens per second, making it one of the fastest and most efficient reasoning models to date.
Introducing K2 Think - a breakthrough in advanced AI reasoning.
— MBZUAI (@mbzuai) September 9, 2025
Developed by MBZUAI’s Institute of Foundation Models and @G42ai, K2 Think delivers frontier reasoning performance at a fraction of the size of today’s largest systems.
Smaller. Smarter. Open to the world.… pic.twitter.com/l3MlOcH30o
In benchmarks, K2 Think leads all open-source models in math reasoning across AIME ’24/’25, HMMT ’25, and OMNI-Math-HARD, placing it among the top-performing systems globally.
“K2 Think has shifted the AI reasoning paradigm from ‘bigger is better’ to ‘smarter is better’. MBZUAI, supported by the UAE ecosystem, is pushing the AI frontier with technology that is open, efficient and highly capable. By proving that smaller, more resourceful models can rival the largest reasoning systems, this milestone marks the beginning of the next wave of AI innovation,” Peng Xiao, Group CEO of G42, said.
Unlike many open models, K2 Think is fully open-source, with released weights, training data, deployment code, and optimisation tools—ensuring transparency and reproducibility for the global AI research community.
“By delivering these advances in a fully transparent framework, we are ushering in a new era of cost-effective, reproducible and accountable AI. For an institution just five years young, we are immensely proud of our global researchers, engineers, and teams who are advancing science and technology with ingenuity and a pioneering spirit," Professor Eric Xing, MBZUAI President and University Professor, added.
K2 Think builds on the UAE’s legacy of open-source AI, following Jais (Arabic), NANDA (Hindi), SHERKALA (Kazakh), and K2-65B, the world’s first fully reproducible foundation model.
Comments ()