Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Gemini Decision: a model and API for choices

Written November 12, 2026

Google's new Decision model and Decisions API are in AI Studio. What the docs say it does, what it costs, and where the limits sit.

What it is

One naming point first, because two products share it: OpenAI has a Decisions API too, covered on the Luna page linked below. This page is Google's. The Gemini API changelog for November 11, 2026 releases Gemini Decision 1 (gemini-decision-1) and the Decisions API for fast, calibrated binary, categorical, and rubric decisions across text, images, audio, video, and PDFs, with zero output tokens.

The Decisions API guide frames the job plainly: when code needs a judgment to branch on, generating prose and parsing it adds latency and brittle parsing logic. A call sends one shared input and a map of questions, and returns one typed answer per question. The model scores the options in a single forward pass instead of writing answer text token by token, and every answer carries calibrated probabilities your code can branch, sort, and route on directly. Google's named uses are tool selection, policy checks, model routing, and context grading.

The three question types

Questions are independent: each one runs against the shared input on its own, questions cannot see each other, and adding or removing one does not change the others. The guide's advice is to keep each question narrow and single-purpose, write the full question in its instructions, and combine the answers with ordinary code. The confidence score is the routing signal: it tells your code whether to act on an answer or send the case to a fallback or a person.

Price and speed

The pricing page lists Gemini Decision 1 at $0.10 per million input tokens on the paid tier, free of charge on the free tier, with no output price because the model generates zero output tokens. The shared input is billed once per request no matter how many questions you ask; each question adds only its own instruction and criteria tokens. Google's launch post reports 150 to 250 milliseconds of end-to-end latency, with median model compute of 50 to 100 milliseconds across inputs up to 16K tokens, staying nearly flat from 1 to 100 questions per request. Those speed figures are Google's, stated as Google's.

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: OpenAI Decisions API: Luna answers yes or no · Gemini 4 Carbon is here: $2 in, $10 out