Overview
Gemini Decision 1 is Google's model for classifying inputs against options or rules you supply. Google released it with the Decisions API on November 11, 2026. For support teams, a practical use is assigning a ticket to an existing queue and sending uncertain cases to a person. This guide is based on Google's documentation.
What Changed
- Pick a category. A
choicequestion returns the selected option, the probability for each option, and a confidence score. - Check a condition or apply a rubric. A yes/no question returns the probability that a condition holds. A rubric question returns a weighted score and the probability of each level.
- Run several checks on one input. A request supports up to 1,000 independent questions. The shared input and all questions together must fit within 262,144 tokens. The input can include text, images, audio, video, and PDFs. Each question sees the shared input and has no access to the other answers.
Compared With a Generative Gemini Call
Use Decision 1 when the choices are known in advance and your app needs a route, a check, or a score. Use a generative call, such as Gemini 4 Hydrogen, when you need a written answer, an explanation, or a suggested resolution. A support system can classify a ticket first and ask a generative model to draft a reply later.
Who Should Use It
Example: Route a Support Ticket
For a fictional ticket that says “I was charged twice for my subscription,” define four queues: Billing, Account Access, Technical Support, and Needs Review. Ask one choice question about the destination. A separate yes/no check can flag mentions of duplicate charges for the reviewer.
Let the application assign the proposed queue only when the result meets a threshold validated on labeled tickets. Send unclear or mixed requests to Needs Review. Start by comparing the proposed routes with human labels, then choose a threshold based on the cost of misrouting and the number of tickets your team can review.
Availability, Pricing, and Limits
Google lists a free tier and a Standard paid rate of $0.10 per million input tokens. Shared input is billed once per request, while each question adds tokens for its instructions and criteria. The API generates zero output tokens and does not support the Batch API.
Google reports median end-to-end latency of 150–250 ms in its tests on inputs up to 16K tokens. Measure latency on your own ticket mix before setting a service target.
