RAIL: Rethinking Auditory Intelligence in Large Audio-Language Models with a CHC-Grounded Benchmark

The article introduces RAIL, a novel evaluation paradigm for large audio-language models (LALMs) based on the Cattell-Horn-Carroll (CHC) cognitive framework, which emphasizes auditory cognition through five core capabilities. RAIL aims to address the limitations of current task-centric evaluation methods by providing structured tasks that assess how models process, retain, and integrate auditory information. This new benchmark reveals significant disparities in the cognitive abilities of 26 evaluated state-of-the-art LALMs, highlighting the need for a more comprehensive understanding of auditory intelligence in AI systems.