AI agents · OpenClaw · self-hosting · automation

Quick Answer

What Is the Gemini Decisions API? Gemini Decision 1 Explained

Published:

The short answer

The Gemini Decisions API is Google’s endpoint for fast, cheap, calibrated decisions — classification without text generation. Launched November 11, 2026, it runs a new model, Gemini Decision 1 (gemini-decision-1), that scores candidate answers in a single forward pass. You send one input and up to 1,000 questions; you get back probabilities and a confidence score for each, in 150–250 ms end-to-end, at $0.10 per million input tokens and $0 for output.

How it works

Each request has a shared input — text, JSON, images, audio, video or PDFs, up to a 262,144-token context window — and a map of questions. Every question is one of three types:

TypeWhat it doesReturns
noul (yes/no)Probability that a statement holds; optional true/false descriptionsnoul between 0.01 and 0.99, confidence
choicePicks one of 2–255 optionsWinning option, probability per option (sums to 1), confidence
scoreGrades against an ordered rubric starting at 0Weighted score, per-level probabilities, confidence

Questions run in parallel against the same input and cannot see each other, so Google recommends narrow, atomic questions combined with ordinary if statements in your code. The confidence score is meant for routing: act automatically when it is high, send to a fallback model or a person when it is low. Calls use client.decisions.create(...) in the google-genai SDK 3.2 or later, or POST /v1beta/decisions over REST.

Pricing and limits

  • $0.10 per 1M input tokens, $0 output — no thinking tokens, no separate audio transcription or OCR charges.
  • Shared input billed once per request, however many questions you ask.
  • No-cost tier in Google AI Studio for prototyping.
  • Gemini Developer API (v1beta) only at launch; support for Gemini in Google Cloud is “coming soon”.
  • Not supported: context caching, Batch API, Flex and Priority inference — repeated input across requests is billed in full.
  • 1–1,000 questions per request; 2–255 options per choice question.

Google says that across its shared benchmark suite the Decisions API is nearly 5x cheaper than running Gemini 3.5 Flash-Lite through the Interactions API, and that Gemini Decision 1 has the lowest calibration error of the models it compared (vendor figures).

What people use it for

Google lists its own deployments: Gemini Notebook routes each query to the right model tier and thinking level (up to 40% lower latency on everyday queries), Stitch applies voice and tap edits to wireframes almost instantly, CC picks emoji reactions from more than 200 options, and Google Search uses it as a safety layer for MCP tools. For builders the obvious jobs are routers in front of expensive models, guardrails and policy checks, ticket triage, document and claims checks that mix a PDF with photos, and rubric grading in evals.

How it compares

Decision models became a category in autumn 2026: OpenAI shipped a Decisions API, and Microsoft and Liquid released decision models — see OpenAI Decisions API vs Microsoft-Decision-1 vs Liquid d1. Google’s version stands out for native multimodal input (audio, video and PDF in the same request) and for pricing that ignores output entirely. Integrations are rolling out in OpenRouter, LangChain, the Vercel AI SDK, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode. For model routing in general, see how to choose an AI model with routing and our current API prices.

Last verified: November 16, 2026.

Sources