ABU DHABI (ALETIHAD)
The Institute of Foundation Models (IFM), established by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), released a fleet of six open-source AI models on Thursday.
In a statement, IFM said the new models, grouped under the name K2 Horizon, are fully open, allowing researchers and developers to inspect, reproduce and adapt them for their own work.
The models range from 0.9 billion to 375 billion parameters and are being released with their model weights, code, training data, methodologies and intermediate development material. Parameters are the internal variables through which an AI model learns patterns and produces responses.
The six models are designed for different computing environments, ranging from small devices such as watches and mobile phones to laptops, on-premise servers and enterprise systems.
The smallest model, K2 Horizon 0.9B, can operate in computing environments with limited memory and power, including watches and smart glasses. The 3.7B and 7B models are intended for mobile phones and other on-device applications.
The 32B model is designed for intensive use on laptops and locally operated servers. The 36B model uses a system that activates only four billion of its parameters when processing a task, which IFM said makes it more cost-effective for local deployment.
The 375B model, of which 23 billion parameters are active at a time, is the flagship of the range and is intended for complex reasoning and enterprise applications.
The six models share a core architecture, vocabulary, training methodology, interfaces and deployment tools, although the smallest model has a more limited vocabulary. This common structure allows developers to test an application on a smaller model and subsequently move to a larger one without changing the deployment process, the institute said.
“Open source is much more than open weights. Science works when others can see the data, follow the method, reproduce the result and improve on it,” said Eric Xing, founder of IFM and President and University Professor of MBZUAI.
“K2 Horizon delivers on that need. Every model in the fleet ships with its training data, recipe and evaluations. This is open science, and we believe it’s the best path forward for AI,” Xing added.
IFM said K2 Horizon incorporates a “diffusion distillation” technique that generates blocks of text units simultaneously. According to the institute, this makes the models operate about three times faster without reducing the quality of their responses.
The models also use what IFM calls a “mixture of value attention” architecture, which it said improves reasoning without requiring additional computing power. A dynamic routing system directs individual tasks to the most cost-effective model in the range.
“Rather than release a single model, we’re releasing an entire fleet at once — six models, from one small enough to run on a watch to a flagship built for enterprise reasoning,” said Hector Liu, Director of IFM’s Silicon Valley lab.
“Every model is built to compete with the best open models at its size, and every one ships with the weights, code, training data and methodology behind it. Developers can prototype on the smallest model, scale to the flagship and verify every claim we make along the way,” Liu added.
IFM is a global AI research institute launched by Abu Dhabi-based MBZUAI in May 2025. It employs researchers, scientists and developers across facilities in Abu Dhabi, Silicon Valley and Paris.
The institute focuses on the open and independent development of advanced foundation models. Its portfolio includes the K2 series, Arabic-language model Jais and PAN, a model designed for physical-world simulation and embodied reasoning.