MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities
Modus tests whether a single decoder-only model can handle arbitrary inputs and outputs across modalities.
The paper frames any-to-any modeling as a way to predict any modality from any mix of others without separate heads, losses, or task pipelines. Its claim is that treating every modality symmetrically lets the same model support chained generation and self-checking across generated modalities. The authors report strong out-of-the-box results and competitive performance against specialist and multitask baselines. Materials are described as open-sourced, and the paper is listed as accepted at ICML 2026. ArXiv · AI/CL/LG's note
The paper frames any-to-any modeling as a way to predict any modality from any mix of others without separate heads, losses, or task pipelines. Its claim is that treating every modality symmetrically lets the same model support chained generation and self-checking across generated modalities. The authors report strong out-of-the-box results and competitive performance against specialist and multitask baselines. Materials are described as open-sourced, and the paper is listed as accepted at ICML 2026. ArXiv · AI/CL/LG's note
score 5