AI can understand, reason over, and generate across multiple modalities

Can it correctly connect information across text, images, audio, or video?

Connects information across modalities

Can it reason across more than one modality at once?

Reasons across multiple modalities

MMMU-Pro directly evaluates reasoning across modalities. It provides meaningful evidence of progress on multimodal reasoning.

Can it create useful outputs across text, image, audio, or video?

Generates useful multimodal outputs

Artificial Analysis's text-to-image leaderboard provides direct evidence that models can generate useful outputs across modalities. It is still only one slice of multimodal generation, but it is a meaningful one.