Gemini Decisions API

Gemini Decision 1 · answers with probabilities

Gemini Decisions APIYes/no, choice and score answers with calibrated probabilities.

The Gemini Decisions API runs Google’s Gemini Decision 1 on closed questions. Send text, JSON or images, define the possible answers, and get one answer per question with a probability for every option. Try it in the playground, then call the same request from your code.

  • 3Question types: yes/no, choice, score
  • 1,000Questions in one API request
  • Text + imagesScreenshots, receipts and photos as context

Gemini Decisions API playground

Pick an example or write your own. Each run uses one credit; new accounts get 20.

A receipt photo plus the employee note: category, policy check and approval.

receipt.jpg
Questions · 3/6
Choice
Yes / No
Yes / No
Gemini Decision 1

Run the decision to see one answer per question, with a probability for every option.

curl https://geminidecisionsapi.net/api/v1/decisions \
  -H "Authorization: Bearer $GEMINI_DECISIONS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "gemini-decision-1",
  "input": "Employee note: \"Team lunch with the Harbor Logistics account after the quarterly review.\" Policy: client meals are reimbursable with an itemised receipt that shows the date and a tax number; any alcohol needs manager approval.",
  "images": [
    "data:image/jpeg;base64,<receipt.jpg>"
  ],
  "questions": {
    "category": {
      "type": "choice",
      "instructions": "Which expense category does this receipt belong to?",
      "criteria": {
        "meals": "Restaurants, catering, coffee",
        "travel": "Flights, trains, taxis, hotels",
        "software": "Subscriptions and licences",
        "office": "Office supplies and equipment"
      }
    },
    "itemised": {
      "type": "noul",
      "instructions": "The receipt is itemised and shows a date and a tax number."
    },
    "needs_approval": {
      "type": "noul",
      "instructions": "The receipt includes alcohol, so a manager must approve it."
    }
  }
}'

Three question types in the Gemini Decisions API

Every question has an id, a type and instructions. Gemini Decision 1 answers only from the options you allow.

Yes / No

noul

A statement that is true or false. The answer is the probability that it is true.

"The customer gives a deadline." → 0.97
Choice

choice

One option from a list you define, with a probability for every option.

billing 0.91 · technical_support 0.06 · sales 0.03
Score

score

One level on an ordered scale you define, lowest first.

Loyal · Annoyed · At risk · Leaving → 2.3

What you can build with the Gemini Decisions API

Any step where your code needs one answer from a fixed list, and needs to know how sure the model is.

  • Support routing

    Send each ticket to the right queue and flag urgent ones.

  • Moderation

    Keep, warn or remove comments under your own policy.

  • Agent tool choice

    Pick an AI agent’s next tool from a closed list.

  • Prompt-injection guard

    Screen messages before they reach your LLM.

  • RAG context grading

    Check that a retrieved passage answers the question.

  • Model routing

    Send each prompt to the cheapest model that can handle it.

One Gemini Decisions API call, many answers

POST your context and questions to /api/v1/decisions with your API key. All questions are answered against the same context in one model call, up to 1,000 per request. Output tokens are free.

  1. Add contextPaste a ticket, comment, message or JSON record.
  2. Define the decisionAsk a yes/no, choice or score question with a fixed set of answers.
  3. Read the probabilitiesSee the chosen answer and the probability of every option.
  4. Call the APISend the same request from your application.
Read the API reference
{
  "id": "run_…",
  "object": "decision",
  "model": "gemini-decision-1",
  "answers": {
    "queue": { "type": "choice", "choice": "billing",
      "probabilities": { "billing": 0.94, "technical_support": 0.03, … },
      "confidence": 0.94 },
    "has_deadline": { "type": "noul", "noul": 0.98 },
    "churn_risk": { "type": "score", "score": 2.4,
      "legend": { "0": "Loyal", "1": "Annoyed", "2": "At risk", "3": "Leaving" },
      "probabilities": { "0": 0.01, "1": 0.08, "2": 0.41, "3": 0.5 }, "confidence": 0.5 }
  },
  "usage": { "input_tokens": 212, "output_tokens": 0 }
}
Response shape. The numbers are illustrative; run the playground for real answers.

Gemini Decisions API pricing

Welcome and check-in gifts enter the Credit account. Purchases add input tokens to a separate Token account. The API uses tokens first, then one credit for the whole request if tokens are insufficient. The playground uses one credit first, then actual input tokens. A request never charges both accounts. Output tokens are free.

20free welcome credits for eligible new accounts
$19per month for 20 million input tokens
$0for output tokens
See pricing

Gemini Decisions API guides

Gemini Decisions API FAQ

What is the Gemini Decisions API?

The Gemini Decisions API answers closed questions with Google’s Gemini Decision 1 model. You define the possible answers; it returns the selected answer and a probability for every option instead of generated text.

Is this an official Google product?

No. Gemini Decisions API (geminidecisionsapi.net) is an independent service that runs Gemini Decision 1 and adds a playground, self-serve API keys and free credits to start. It is not affiliated with or endorsed by Google.

Which question types are supported?

Three: noul (yes/no, returns the probability that the statement is true), choice (one option from a list you define) and score (one level on an ordered scale you define).

Can the Gemini Decisions API read images?

Yes. Attach up to 4 PNG or JPEG images per API request (3 in the playground) together with text or on their own: receipts, screenshots, product photos and documents. Each image gets a key such as image_1 that the context or a question can refer to.

How many questions can one request ask?

Up to 1,000 questions per API request, all answered against the same context in one model call. The playground runs up to 6 questions at a time.

How is usage billed?

Playground runs use one welcome credit first. API calls use purchased input tokens first. All questions in a request share one model call, so input tokens grow with the context and question text. Output tokens are free.

About this service

Gemini Decisions API (geminidecisionsapi.net) is an independent service that runs Google’s Gemini Decision 1 model. It is not affiliated with, sponsored by or endorsed by Google. “Gemini” is a trademark of Google LLC and is used here only to identify the model.