Gemini Decisions API: Fast, Calibrated Multimodal Decisions
Google introduces the Gemini Decisions API, powered by Gemini Decision 1, for fast, calibrated multimodal decisions. Evaluate up to 1,000 questions across text, images, audio, video, and PDFs in a single pass with confidence scores. The API offers low latency and cost-effective routing for developers.
Tags
Quick summary
Google introduces the Gemini Decisions API, powered by Gemini Decision 1, for fast, calibrated multimodal decisions. Evaluate up to 1,000 questions across text, images, audio, video, and PDFs in a single pass with confidence scores. The API offers low latency and cost-effective routing for developers.
Gemini Decisions API: Fast, Calibrated Multimodal Decisions
Google has introduced the Gemini Decisions API, a new endpoint designed to facilitate fast, probabilistic choices across diverse data types. Powered by the Gemini Decision 1 model, this tool allows developers to evaluate up to 1,000 independent questions in a single pass. The API supports text, images, audio, video, and PDFs, enabling applications that require rapid grading, policy checks, or context evaluation without the latency associated with traditional generative text decoding.
According to Google, the primary advantage of the Decisions API lies in its ability to compute confidence scores for multiple questions simultaneously. Instead of generating answer text token by token, Gemini Decision 1 scores candidate options in a single forward pass. This architecture returns probabilities and confidence scores with an end-to-end latency of 150-250 ms. The pricing structure is designed for cost efficiency, charging $0.10 per 1M input tokens while billing zero output tokens.
Model Architecture and Performance
The Gemini Decision 1 model (gemini-decision-1) is built on Gemini's native multimodal architecture. This foundation allows the model to process text, images, audio, video, and PDFs directly within a single forward pass. Users can combine modalities in a shared input, such as pairing a call recording with account metadata or a scanned PDF claim with damage photos. Furthermore, the model allows users to tune resolution on image and video inputs to control visual detail and token cost.
Google reports that Gemini Decision 1 scores options without token-by-token decoding. This approach keeps the median latency of model compute between 50-100 ms, with the total end-to-end latency remaining between 150-250 ms. The performance remains nearly flat even as the number of questions per request increases from 1 to 100. The model supports contexts up to 16K tokens and can handle inputs up to 256K tokens.
In terms of cost-effectiveness, Google states that Gemini Decision 1 with the Decisions API is nearly 5x cheaper across a shared benchmark suite compared to Gemini 3.5 Flash-Lite with the Interactions API. This significant reduction in cost is attributed to the lack of output token billing and the efficiency of the single-pass scoring mechanism.
On benchmarks spanning agent routing, reasoning, ranking, and multimodal understanding, Google claims that Gemini Decision 1 leads across the suite with the lowest calibration error. This strong performance and calibration are central to the API's value proposition for applications requiring high reliability in decision-making.
Decision Types and Evaluation
The API provides purpose-built primitives for bounded choices, including tool selection, policy checks, model routing, and context grading. Users can evaluate multiple questions against a shared multimodal input in a single request. The system supports three distinct decision types, each returning a confidence score from 0 to 1.
- Yes/No (noul): This type estimates the probability that a yes/no condition holds. It supports optional true and false description criteria to refine the evaluation.
- Multiple Choice (choice): This type selects from 2 to 255 choices. It returns the winner and normalized probabilities for every candidate option.
- Ordered Rubric (score): This type grades inputs across an ordered rubric starting at 0. It returns a continuous weighted score and per-level probabilities.
Every answer generated by the API includes a confidence score. This feature allows application code to route ambiguous cases to a fallback system or a human reviewer, ensuring that low-confidence decisions do not proceed automatically.
Practical Implementation
Developers can start building with the Decisions API by accessing an API key in Google AI Studio. The input field utilizes the same Content block format as the Interactions API. Google provides a Python example demonstrating how to check a photo of a breakroom counter, combined with a text note, against all three question types in one request.
The example below illustrates how to define questions for spill detection, cleanup actions, and severity rating:
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 counter check 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 action fits the scene?",
"criteria": {
"wipe_with_cloth": "Small counter spill with a cloth or towel already nearby.",
"dispatch_janitorial": "Large floor hazard, broken glass, or active leak.",
"ready_for_meeting": "Counter is clean and ready.",
},
},
"severity": {
"type": "score",
"instructions": "Rate the severity of the counter mess.",
"criteria": [
"Clean and ready.",
"Minor localized spill.",
"Major spill or safety hazard.",
],
},
},
)Enterprise and Internal Use Cases
Google highlights several internal teams that are already leveraging the Decisions API as a low-latency control plane. These implementations demonstrate the versatility of the tool in production environments.
- Dynamic Model Routing in Gemini Notebook: Gemini Notebook uses the Decisions API as a real-time router to direct incoming queries to the optimal model tier and thinking level. The API classifies intent in milliseconds, serving everyday queries with up to 40% lower latency while reserving deep reasoning models for complex tasks.
- Rapid Design Iteration in Stitch: The rapid canvas experiment in Stitch uses the API to generate playable wireframes of a product flow in seconds. It applies edits made by voice or tap almost instantly, handing off screens to the full Stitch experience once the flow is finalized.
- Emoji Selection in CC: CC from Google Labs uses the API to pick the right emoji reaction from a palette of more than 200 options in a fraction of a second, ensuring reactions appear instantly without adding latency.
- Safety Layer for MCP Tools in Search: Google Search utilizes the Decision API to detect adversarial behavior and content against adversarial evals with high efficacy within the latency envelope.
Getting Started and Ecosystem
Developers can begin prototyping immediately by accessing the API key in Google AI Studio. Google offers demo apps such as Doorstep Parcels, Speed Sketch, and Find the Zoo Keeper to help users explore the capabilities. Documentation is available for Python, JavaScript, and REST guides.
Ecosystem integrations are rolling out across various platforms, including OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit, and OpenCode, alongside the google-genai SDKs. Support is coming soon to Gemini in Google Cloud.
The API includes an API no-cost tier for prototyping. Pricing remains at $0.10 per 1M input tokens with no charges for output tokens. There are no variable thinking-token bills or separate audio transcription and OCR costs. The shared input is billed once per request, regardless of the number of questions evaluated, as each question adds only its own instruction and criteria tokens.
For those interested in further development, Google encourages signing up for newsletters to receive product updates, event information, and special offers. The API represents a significant step forward in providing calibrated, fast, and cost-effective multimodal decision-making capabilities for developers.
Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api



