Google launches multimodal Gemini Decision 1 API

The decision model category now has three named entrants: Jev, OpenAI, and Google. Gemini Decision 1 copies the question vocabulary directly, reusing the term "Noul" for a yes/no question, which suggests the interfaces are converging on a shared shape. What differs is the input surface: Jev is text and JSON only, OpenAI handles text and images, and Gemini Decision 1 covers text, images, audio, video, and PDFs. In a test on a video of a stomping robot in Yonkers, the model returned probabilities for four yes/no questions, including 0.9234 for a robot and 0.9230 for the USA. Jev and OpenAI return exact 1.0 or 0.0 values for the same style of question, so the practical difference is whether a caller receives a graded confidence or a hard label, a distinction that may matter in pipelines where downstream logic needs to know how sure the answer is.