Skip to content

AI Agents & Coding · Freemium

Gemini Decisions API

The Gemini Decisions API is an endpoint in Google's Gemini Developer API for fast, closed-set decisions. Instead of generating text, it uses a dedicated model, Gemini Decision 1 (model ID gemini-decision-1), to score a fixed set of possible answers and return calibrated probabilities and a confidence value for each question. Google DeepMind launched it on November 11, 2026, through Google AI Studio and the v1beta Gemini API.

What it does: a request contains one shared input and between 1 and 1,000 independent questions. The input can be text, images, audio, video or PDF documents, up to a 262,144-token context window shared by the input and all questions. Each question has one of three types. A yes/no question (called "noul" in the API) returns the probability that a statement holds, between 0.01 and 0.99. A choice question picks one of 2 to 255 options and returns a probability for every option. A score question rates the input against 2 to 255 ordered rubric levels and returns a probability-weighted score. Questions are evaluated in parallel and cannot see each other, so adding one does not change the others.

Who it is for: developers building agents and applications that need many small, bounded judgments: routing a request to a model tier, queue or tool; checking inputs and outputs against many policy rules at once; grading or ranking candidate results against a rubric; and reacting to images, audio or video in real time. Google says its own products use it for model routing in Gemini Notebook, emoji selection in CC and a safety layer for MCP tools in Search.

Key features: - Zero output tokens: only input tokens are billed, and the shared input is billed once per request regardless of how many questions are asked. - Calibrated probabilities and a confidence score on every answer, designed for thresholding in code. - Native multimodal input with adjustable media resolution for images and video. - Google cites 150 to 250 ms end-to-end latency, staying nearly flat from 1 to 100 questions per request. - Python and JavaScript support in the google-genai SDK 3.2 or later, plus REST at /v1beta/decisions.

Limitations: no context caching, no Batch API, no Flex or Priority inference, text-only question instructions, no conversation-history or tool-call input blocks, and no YouTube or gs:// URIs. It is available on the Gemini Developer API only; Google says Gemini in Google Cloud support is coming.

Pricing type: freemium. There is a free tier for prototyping in Google AI Studio (200 requests per minute and 10,000 requests per day). The paid tier costs $0.10 per million input tokens with no output charge, as of November 12, 2026.

Alternatives: structured output from small generative models such as Gemini 3.5 Flash-Lite or Claude Haiku 5.5, logprob-based classification with other LLM APIs, or a custom fine-tuned classifier.

Official URL: https://aistudio.google.com/docs/decisions-api

Official site