Google's newest model has one job, and it never writes a word doing it. Decision 1 reads text, images, video, audio and PDFs in a single forward pass, up to 262,144 input tokens per request.