Meow-Omni 1: a multi-modal feline LLM (arxiv.org)

Researchers have introduced Meow-Omni 1, the first open-source quad-modal Large Language Model (LLM) designed specifically for feline ethology. This innovative model addresses the challenge of deciphering animal intent, particularly in cats, by integrating video, audio, physiological time-series data, and textual reasoning—all of which enhance understanding beyond basic behavioral matching. With its unique architectural adaptations and specialized scientific encoders, Meow-Omni 1 achieves a state-of-the-art intent-recognition accuracy of 71.16% on the newly developed MeowBench benchmark, significantly surpassing existing vision-language and omni-modal models. The significance of Meow-Omni 1 lies in its potential to bridge the gap between technology and animal behavior, providing a scalable framework for understanding inter-species communication. The open-source release of its complete pipeline, including model weights, training framework, and the Meow-10K dataset, aims to advance AI applications in real-world settings, such as veterinary diagnostics and wildlife conservation. This breakthrough not only enhances our ability to interpret feline intent but also sets a precedent for similar approaches in studying other animal behaviors using advanced AI methodologies.