Demographic and Linguistic Bias Evaluation in Omnimodal Language Models

ArXi:2604.10014v1 Announce Type: cross This paper provides a comprehensive evaluation of graphic and linguistic biases in omnimodal language models that process text, images, audio, and video within a single framework. Although these models are being widely deployed, their performance across different graphic groups and modalities is not well studied. Four omnimodal models are evaluated on tasks that include graphic attribute estimation, identity verification, activity recognition, multilingual speech transcription, and language identification.