ARIA: NATIVE MULTIMODAL MOE

Summary from the TLDR AI newsletter

An extremely impressive new model with open weights, code, and research report for vision and language. It matches closed models for long video understanding, which has eluded open models like Pixtral and Molmo.