UAE unveiled the artificial intelligence of K2 Think; OpenAI and Dipsic

Emirates University of Artificial Intelligence (MBZUAI) from a new, low -cost rational model K2 Think Unveiled. This model is developed in collaboration with the UAE G42; According to the creators, the model, despite a much smaller size, offers a performance at the level of flagship and large -scale lead companies such as Openai and Deepseek.

According to the CNBC report, the most amazing feature of K2 Think is its size. This model is just 2 billion parameters It seems very small compared to the Chinese R1 model Deepseek with 2 billion parameters. However, the G42 researchers claim that K2 Think in difficult benchmarks related to mathematics, coding and science (such as Aime24 and GPQA-Diamond) has performed equal or even better than its competitors.

“What we are discovering is that much more can be done with less opportunities,” says Richard Morton, CEO of the Institute for Basic Models at MBZUAI. Of course, this model is based on the QWen 2.5 Alibaba text model and is developed and tested on Cerebas hardware.

Artificial Intelligence Model K2 Think Emirates

According to the makers, the high performance of K2 Think has been obtained through a combination of advanced methods. These methods include carefully adjusted with the Long Chain-Oph-Though “and” Test-Time Scaling “, which improves performance by allocating more computational resources at the moment of inference.

“What makes our model special is that we treat it more like a system than just one model,” the researchers say. “This is a system work in which all of these methods are combined and provide the final result.”

K2 Think developers emphasize that their aim was not to build another chat with ChatGpt. Instead, this model is designed for specific applications in areas such as mathematics and science that require deep reasoning. They say that with this specific application, this model is deeply compressed, instead of having five or five people five years to think about a specific question or examine a set of clinical tests.