Google introduced the Gemini Decisions API on November 11 for developers who need a system to choose among defined options. The endpoint uses Gemini Decision 1 to score independent questions about one shared input. That gives an application confidence scores it can use for routing, grading or a separate review step.
What the endpoint returns
A request can mix yes-or-no checks, multiple choice and ordered rubric scores. Google says the input can combine text, images, audio, video and PDFs. The endpoint accepts up to 1,000 independent questions and shared input of up to 256K tokens. Its answers include confidence scores between zero and one, which developers can use to decide when to route uncertain cases to another step.
What Google measured and charges
Google reports median latency of 150 to 250 milliseconds end to end in the specific tests it describes, across contexts up to 16K tokens. Its comparison covers one to 100 questions in a request. The company lists pricing of $0.10 per million input tokens and zero output-token charges, with a no-cost tier in Google AI Studio for prototyping. The shared input is billed once per request, while question instructions add input tokens.
How to decide whether it fits
The API is designed for questions where an application can define the choices ahead of time. That fits decisions such as which tool to use or whether an input satisfies a criterion. A developer still needs to decide what an acceptable confidence score is for each workflow. Test the actual input length, write examples of ambiguous cases and measure how often those cases should go to a person or another model.
