While previous embedding models were largely restricted to text, this new model natively integrates text, images, video, audio, and documents into a single numerical space — reducing latency by as muc ...