Google released Gemini Decision 1 through its Gemini API and AI Studio. Developers provide a defined set of choices and receive probabilities rather than an unrestricted written answer. The model accepts several kinds of input and can evaluate multiple questions together. Applications can use its confidence scores to send uncertain cases to another model or a human reviewer.
Google opened Gemini Decision 1 through the Gemini application programming interface, or API, and Google AI Studio on November 11.[1], [2]
The model chooses within boundaries supplied by an application: developers give it candidate answers and decision criteria. It returns probabilities for those choices instead of composing an open-ended response. One shared input can support several questions in the same request.[1]
The interface supports yes-or-no checks, selection among candidate answers and scoring against an ordered rubric. Each answer includes a confidence measure, which developers can use to refer uncertain cases to another model or a person. Inputs can combine text, images, audio, video and documents. Google describes uses including tool selection, policy checks and choosing which model should handle a request.[1]
Google says Gemini Notebook uses the model to choose a model and reasoning level. Other company examples involve interface design, emoji selection and safety checks for search tools. These are deployments described by Google. Integrations with development tools are being introduced alongside the API release. Access through Google Cloud remains a planned extension, separate from the interface already opened through AI Studio and the Gemini API.[1]
