Falcon-OCR and Falcon-Perception

The Technology Innovation Institute (TII) in Abu Dhabi has expanded its influential Falcon family of open-source AI models with two new multimodal entrants: Falcon-OCR and Falcon-Perception. These models mark a significant step in bringing advanced vision-and-language capabilities to the open-source community, directly challenging proprietary offerings. Falcon-OCR is specifically fine-tuned for Optical Character Recognition (OCR) tasks, enabling it to accurately read and interpret text from a wide variety of images, from documents to real-world scenes. Falcon-Perception, meanwhile, is a more general-purpose Vision Language Model (VLM) designed to understand and reason about visual content, answering questions about images and describing scenes.

Both models are built upon TII's established Falcon architecture, known for its strong performance in language tasks, and are now available for download and experimentation on the Hugging Face platform. In a move that underscores the collaborative nature of the open-source AI ecosystem, community developers have already submitted a pull request to integrate these new Falcon models into the popular llama.cpp project. This integration would allow users to run the vision-capable Falcons efficiently on consumer hardware, using CPU and GPU acceleration for local, private inference, significantly lowering the barrier to entry for advanced multimodal AI.