Gemini Decision 1 is Free on Google AI Studio

Google AI Studio’s Decisions API uses Gemini Decision 1 to turn model judgments into typed results that software can act on, rather than generating prose that code must parse. A request supplies one shared input and one or more questions; each question is evaluated independently in a single forward pass, keeping latency nearly flat as questions are added. The API emits no text output tokens, and returns calibrated probabilities plus a confidence value to help applications decide whether to act automatically or escalate. It is presented as free to use, and can be called through Python or JavaScript SDKs or REST, using SDK version 3.2 or later.

The API supports three question formats: yes/no checks, which return the probability that a statement is true; fixed-choice classification, which returns the leading option and probabilities for all options; and rubric-based scoring, which rates input across ordered levels and calculates a probability-weighted score. The documentation advises writing narrow, atomic checks and combining their results in ordinary application code. It also highlights practical safeguards: define criteria when a yes/no question is ambiguous, set thresholds within the yes/no probability range of 0.01-0.99, and handle choice results cautiously because probabilities are always distributed across listed options even when none fits. A catch-all category or separate scope check can reduce misrouting, while confidence thresholds can flag uncertain cases.