Google has launched a “Gemini Decisions API” powered by a new model, Gemini Decision 1, aimed at fast “closed-set” decisions. The API evaluates up to 1,000 yes/no, multiple-choice, or rubric-scoring questions in a single request and returns calibrated probabilities rather than free-form answer text. It takes a shared multimodal input that can include text plus media such as images, audio, video, and PDFs.
The model is described as similar in structure to earlier “decision model” offerings and reuses terminology for yes/no questions. Across the reports, Gemini Decision 1 is positioned for applications like tool selection, policy checks, model routing, and context grading, where software can use probabilities and confidence scores to decide what to do next. One report includes a test example for yes/no questions against a video input, showing probabilities that are not limited to exact 0.0 or 1.0.
Pricing is consistently reported as $0.10 per million input tokens, with output tokens billed at zero cost. Jev is cited as cheaper among the listed decision-model options in one account, while another report highlights Google’s claimed latency and calibration performance and references a no-cost tier in Google AI Studio.
