Abu Dhabi’s IFM open-sourced six variants of the K2 Horizon model, covering sizes from 0.9 billion to 375 billion parameters.

Beating AI News Flash: The Institute for Foundation Models (IFM) under the UAE’s Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has launched K2 Horizon, open-sourcing six models at once: 0.9B, 3.7B, 7B, 32B, 36B-A4B, and 375B-A23B. The 0.9B model is targeted at edge devices like smartwatches and glasses, while the 7B model can run on mobile phones. The largest 375B-A23B uses a Mixture of Experts (MoE) architecture, activating around 23 billion parameters per generated token. The institute stated that the 0.9B, 3.7B, and 7B models have set new benchmark records for their respective sizes. Beyond final model weights, training code, data or data construction recipes, intermediate checkpoints, training logs, and evaluation records are also made public. Some data is subject to license restrictions and cannot be redistributed directly, though its sources, construction methods, and ratios will be disclosed. The models and code are released under the Apache 2.0 license. Artificial Analysis scored the 375B-A23B an Intelligence Index of 47, close to MiniMax-M3’s 45. The model performs stronger on agent tasks but is relatively weaker in knowledge and complex reasoning.