Google launches Gemini Decisions API and Gemini Decision 1

Google has launched the Gemini Decisions API and its new Gemini Decision 1 model. It is for closed-set work such as model routing, policy checks, tool selection, and context grading, rather than free-form text generation. Google says the endpoint is available through the Gemini API in Google AI Studio.

The API accepts text, images, audio, video, or PDFs as shared input and can evaluate up to 1,000 independent questions in a request. Its question types are yes/no, multiple choice, and ordered rubric scoring; each answer includes a probability and confidence score so low-confidence results can be routed to a fallback or a human reviewer. Google describes the model as scoring options in a single forward pass instead of decoding output token by token.

Google lists Gemini Decision 1 at $0.10 per million input tokens, with no output-token charge, and says an API no-cost tier is available in Google AI Studio for prototyping. This dedicated decision category is relevant to teams separating fast classification or routing from generative model calls, but reported latency, calibration, and benchmark results still need validation against a team's own inputs, thresholds, and failure handling.

Start with a reversible, non-critical route or content check. Log confidence, error distribution, end-to-end latency, and actual token cost before expanding use. A probability score is not a safety guarantee: permission, compliance, and high-risk tool workflows still need independent rules, approvals, and human fallback.

Related catalog entry