Gemini Decision 1 (gemini-decision-1) is a Google DeepMind model released on November 11, 2026 as the engine of the Gemini Decisions API. Unlike Gemini chat models, it does not generate text. A developer sends one input and a set of typed questions, and the model returns a probability, a selected option or a score for each question, together with a confidence value between 0 and 1. Google describes it as a model for fast, calibrated decisions such as routing, policy checks and grading. See our launch report on the Gemini Decisions API.

How it works

Google says the model scores the candidate options for every question in a single forward pass rather than writing an answer token by token. Each request supports one to 1,000 questions of three types:

  • noul: a yes/no question that returns the probability of “yes”.
  • choice: picks one of 2 to 255 options and returns probabilities that sum to 1.
  • score: rates the input on 2 to 255 ordered levels (2 to 10 recommended) and returns a probability-weighted score.

Input can mix text, images, audio, video and PDF documents, up to a context window of 262,144 tokens shared by the input and all questions. Media can be passed inline, through the Files API or by HTTPS URL; YouTube links and direct gs:// URIs are not accepted. Conversation history has to be serialized as text.

Performance

Google reports end-to-end latency of 150 to 250 milliseconds and a median of 50 to 100 ms of model compute, nearly flat from 1 to 100 questions per request. It says the model leads its agent-routing, reasoning, ranking and multimodal benchmark suites with the lowest calibration error, but it has published those results only as charts, and they have not been independently reproduced. Google uses the model in Gemini Notebook for model-tier routing, where it reports up to 40% lower latency on everyday queries, and as a safety layer for MCP tools in Google Search.

Pricing and availability

The paid tier costs $0.10 per million input tokens for every modality; output is never billed because the model produces no text tokens. The shared input is billed once per request regardless of the number of questions. Google says this makes it nearly five times cheaper than Gemini 3.5 Flash-Lite across its benchmark suite. The model is available through the Gemini Developer API (v1beta) and Google AI Studio, which has a free tier. Context caching, the Batch API and Flex or Priority inference are not supported, and support for Gemini in Google Cloud is listed as coming soon.

Alternatives

Competing decision models include OpenAI’s Decisions API, which runs on gpt-6-luna and is also priced at $0.10 per million uncached input tokens, and TypeSafe AI’s Jev, listed at $42 per billion input tokens. Cloudflare’s open Clef models are covered in our Clairic extraction model report. For text generation, Google’s frontier option is Gemini 4 Argon.