Gemini Decisions API

by Adi Gunawan 11 November 2026 6 min read 9x views

Introducing the Gemini Decisions API, powered by our new Gemini Decision 1 model. Evaluate up to 1,000 yes/no, choice, and rubric-scoring questions in a fast, sole continue throughout text, images, audio, video, and PDFs.

Min Choi

Product Manager, Google DeepMind

Felix Yu

Principal Scientist, Google DeepMind

Your browser does not assistance the audio element.

Listen to article

[[duration]] minutes

This satisfied is generated by Google AI. Generative AI is experimental

Today, we’re introducing the Gemini Decisions API, a new endpoint for fast, closed-set decisions powered by our new Gemini Decision 1 example (gemini-decision-1), accessible through the Gemini API in Google AI Studio.

The Decisions API gives you purpose-built primitives for bounded choices akin tool selection, guideline checks, example routing, and environment grading. You can continue one multimodal input throughout text, images, audio, video, and PDFs (up to 256K tokens) alongside alongside up to 1,000 autonomous questions. Rather than generating answer content token by token, Gemini Decision 1 scores your applicant options in a sole onward pass, returning probabilities and confidence scores in 150-250 ms end-to-end latency at $0.10 per 1M input tokens alongside zero output tokens billed.

Comparison of Gemini Decision and another APIs against assorted benchmarks

Strong achievement and calibration

On benchmarks spanning delegate routing, reasoning, ranking, and multimodal understanding, Gemini Decision 1 leads throughout the suite, alongside the lowest calibration error.

Comparison of the Decisions API and another APIs to benchmark

Gemini Decision 1 scores options without token-by-token decoding, keeping median latency 50-100 ms of example compute (150-250 ms end-to-end), throughout multimodal inputs and contexts up to 16K tokens, remaining nearly flat from 1 to 100 questions per request. With zero output tokens billed, Gemini Decision 1 alongside the Decisions API is nearly 5x cheaper throughout the shared benchmark suite than Gemini 3.5 Flash-Lite alongside the Interactions API.

Native multimodal evaluation throughout three decision types in one pass

Built on Gemini’s native multimodal architecture, Gemini Decision 1 can procedure text, images, audio, video, and PDFs immediately in a sole onward pass. You can merge modalities in one shared input, specified as pairing a call record alongside document metadata or a scanned PDF assertion alongside damage photos, and tune resolution on depiction and video inputs to authority ocular item and token cost.

You can measure multiple questions against that shared input in a sole request, mixing yes/no checks, multiple choice, and rubric scores. Every answer includes a confidence mark (from 0 to 1) so your code can path ambiguous cases to a fallback or individual reviewer:

noul (Yes/No): Estimates the probability that a yes/no circumstance holds, alongside optional true and false clarification criteria.
choice (Multiple Choice): Selects from 2 to 255 choices, returning the victor and normalized probabilities for all candidate.
score (Ordered Rubric): Grades inputs throughout an ordered rubric (starting at 0), returning a uninterrupted valued mark and per-level probabilities.

The input site uses the identical Content obstacle format as the Interactions API. Here is how to inspect a photo of a breakroom counter, together alongside a content note, against all three inquiry types in one request:

from google import genai client = genai.Client(http_options={"api_version": "v1beta"}) result = client.decisions.create( model="gemini-decision-1", input=[{ "type": "image", "uri": "https://storage.googleapis.com/generativeai-downloads/images/spill.jpg", "mime_type": "image/jpeg", }, { "type": "text", "text": "Breakroom oppose inspect before the 9 AM client meeting.", }, ], questions={ "has_spill": { "type": "noul", "instructions": "Is there an uncleaned liquid spill on the counter?", }, "next_action": { "type": "choice", "instructions": "Which cleanup act fits the scene?", "criteria": { "wipe_with_cloth": "Small oppose spill alongside a cloth or towel already nearby.", "dispatch_janitorial": "Large flat hazard, damaged glass, or energetic leak.", "ready_for_meeting": "Counter is spotless and ready.", }, }, "severity": { "type": "score", "instructions": "Rate the severity of the oppose mess.", "criteria": [ "Clean and ready.", "Minor localized spill.", "Major spill or safety hazard.", ], }, }, )

Gemini detected a spill alongside a next_action wipe_with_cloth

How teams are construction alongside the Decisions API

Teams throughout Google are already using the Decisions API as a low-latency authority plane, shipping features merely by defining options fairly than training tradition classifiers or waiting on generative calls:

Dynamic example routing in Gemini Notebook: Gemini Notebook uses the Decisions API as a real-time router to immediately straightforward incoming queries to the optimal example tier and thinking flat for all request. The Decisions API classifies intent in milliseconds to assist mundane queries alongside up to 40% lesser latency during reserving profound reasoning models for complex tasks.
Rapid scheme repeat in Stitch: The new rapid canvas test in Stitch uses the Decisions API to create playable wireframes of a merchandise stream in seconds and to use edits made by sound or tap nearly instantly. When the stream feels right, one click hands the screens off to the complete Stitch cognition for deeper scheme polish.
Emoji choice in CC: CC from Google Labs uses the Decisions API to choice the correct emoji reply from a palette of additional than 200 options in a fraction of a second, so reactions appear immediately without adding latency.
Safety tier for MCP tools in Search: Google Search uses the Decision API to detect adversarial behavior and satisfied against adversarial evals alongside elevated efficacy inside the latency envelope.

Stitch Rapid uses the Decisions API to scheme an complete merchandise stream in seconds (1x speed).

Start construction today

Gemini Decision 1 is priced at $0.10 per 1M input tokens and $0 for output tokens, alongside an API no-cost tier you can use in Google AI Studio for prototyping. Your shared input is billed formerly per petition no matter how many questions you measure (each inquiry adds lone its own education and criteria tokens). There are no output-token charges, changeable thinking-token bills, or distinct audio transcription and OCR costs.

You can commencement construction today by grabbing an API key in Google AI Studio, trying our Doorstep Parcels, Speed Sketch, and Find the Zoo Keeper demo apps, and following the Python, JavaScript, and REST guides in the Decisions API documentation.

Ecosystem integrations are rolling out throughout OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit, and OpenCode alongside the google-genai SDKs, alongside assistance coming soon to Gemini in Google Cloud.