AION-1: Omnimodal Foundation Model for Astronomical Sciences

Kyunghyun Cho (Genentech / NYU) · Miles Cranmer (University of Cambridge) · Alberto Bietti (Flatiron Institute, Simons Foundation) · Ruben Ohana (NVIDIA) · Michael McCabe (Flatiron Institute) · Francois Lanusse (CNRS) · Shirley Ho (Polymathic / Simons / NYU) · Liam Parker (UC Berkeley / Polymathic AI) · Jeff Shen (Princeton University Polymathic AI) · Ollie Liu (University of Southern California) · Tom Hehir (University of Cambridge) · Leopoldo Sarra (Axiomatic AI) · Lucas Meyer (INRIA) · Micah Bowles (Ellison Institute for Technology) · Sebastian Wagner-Carena (New York University / Simons Foundation) · Helen Qu (University of Pennsylvania) · Siavash Golkar (Flatiron Institute) · Hatim Bourfoune (Presight) · Pierre Cornette (IDRIS - CNRS) · Keiya Hirashima (RIKEN / Flatiron Institute) · Geraud Krawezik (Flatiron Institute) · Nicholas Lourie (New York University) · Rudy Morel (Flatiron Institute, Polymathic AI) · Payel Mukhopadhyay (University of California, Berkeley) · Mariel Pettee (University of Wisconsin--Madison)
astronomical objectsclassification taskscross-modal token sequencesfoundation modelsgeneration tasksheterogeneous data typesjoint modelingmasked modelingmodality-specific tokenizationmultimodalobject retrievalregression tasksscalable blueprinttransformer-based modelingtwo-stage architecture

While foundation models have shown promise across a variety of fields, astronomy lacks a unified framework for joint modeling across its highly diverse data modalities. In this paper, we present AION-1, the first large-scale multimodal foundation family of models for astronomy. AION-1 enables arbitrary transformations between heterogeneous data types using a two-stage architecture: modality-specific tokenization followed by transformer-based masked modeling of cross-modal token sequences. Trained on over 200M astronomical objects, AION-1 demonstrates strong performance across regression, classification, generation, and object retrieval tasks. Beyond astronomy, AION-1 provides a scalable blueprint for multimodal scientific foundation models that can seamlessly integrate heterogeneous combinations of real-world observations. Our model release is entirely open source, including the dataset, training script, and weights.