The Technology Innovation Institute (TII), the applied research arm of the Advanced Technology Research Council (ATRC) in Abu Dhabi, has announced the launch of Falcon Perception, a next-generation multimodal artificial intelligence model designed to compete with leading global systems.
Falcon Perception is a compact yet powerful model with around 600 million parameters. Despite its relatively small size, it delivers strong performance in tasks such as object segmentation, visual understanding, and document analysis. The model is designed to match or approach the performance of much larger AI systems while using significantly fewer computing resources.
Multimodal AI refers to systems that can process and understand different types of data at the same time, such as images and text. Unlike traditional AI models that mainly focus on language, Falcon Perception is built to understand the visual world along with text in a single unified architecture.
According to TII, the model can identify and segment objects in complex images, read text from documents, and respond to natural language queries about visual content. For example, users can ask the model to identify a specific object in a crowded image or count items in a scene, and it can accurately locate and highlight them.
This approach removes the need for multiple separate systems for vision and language tasks, reducing complexity and improving efficiency. It also makes the model more suitable for real-world applications where computing resources and speed are limited.
The institute highlighted potential use cases including robotics, industrial automation, manufacturing inspection, and large-scale data labeling for AI training.
TII CEO Dr. Najwa Aaraj said the model reflects the institute’s focus on building practical and advanced AI systems that can be deployed across industries while strengthening sovereign AI capabilities in the UAE.
Falcon Perception has shown strong results in benchmark tests, including object segmentation, complex visual understanding, and document intelligence tasks. It performs close to or on par with larger models such as Meta’s SAM3 and Alibaba’s Qwen systems in key evaluation areas.
Dr. Hakim Hacid, Chief Researcher at TII’s Artificial Intelligence and Digital Research Center, said the model demonstrates that a single efficient architecture can handle complex vision and language tasks without relying on multi-stage systems.
Falcon Perception is the first Falcon model focused specifically on dense multimodal perception tasks. It will be released as open-source on Hugging Face, supporting global research collaboration and further development in multimodal AI systems.
