The text and image encoders within Embed 3 share a unified latent space, significantly reducing complexity for users. Our model is also uniquely performant at mixed modality searches.

→ View original post on X — @cohere