Free guide

Gemini Decisions API: Yes/No, Pick and Score Answers With a Confidence Score

Google's new model doesn't write answers. It picks them, and tells you how sure it is. What it does, how it works, what it costs, and your first call in a few minutes.

By Beau Johnson·
The Short this guide goes with: Gemini Decisions API: Yes/No, Pick and Score Answers With a Confidence Score

Quick answer

The Gemini Decisions API is a new Google endpoint, powered by the Gemini Decision 1 model (gemini-decision-1), that answers yes/no, multiple-choice and rubric-score questions about your input instead of writing text. One request can hold up to 1,000 questions about the same text, image, audio, video or PDF input, and every answer comes back with a confidence score from 0 to 1. It costs $0.10 per million input tokens with nothing billed for output, and there's a no-cost tier in Google AI Studio for prototyping.

The short version

  • Gemini Decision 1 doesn't generate text. It scores your options in a single pass and returns probabilities plus a confidence score for each answer.
  • Three question types: yes/no (noul), pick one of 2 to 255 options (choice), and a score on your own rubric (score). Up to 1,000 questions per request.
  • $0.10 per million input tokens, $0 for output, a no-cost tier in AI Studio, and Google says 150 to 250 ms end to end.

How it actually works

A panel of judges, not a commentator

Most models are commentators: they talk you through an answer word by word, and your code has to read the text. Gemini Decision 1 works like a panel of judges. Every judge looks at the same input and holds up a card at the same time: yes or no, a pick from your list, or a score. Your code reads the cards directly, with no text to parse.

One input, many isolated questions

You send one shared input (text, images, audio, video or PDFs, up to 256K tokens) and a map of up to 1,000 questions. Google's docs say every question runs against the shared input in one forward pass and the questions can't see each other, so latency stays nearly flat as you add questions. The shared input is billed once per request, no matter how many questions you ask.

Confidence tells your code what to do next

Each answer comes with a confidence score from 0 to 1. Google's suggestion: the answer tells your code which action fits, and the confidence tells it whether to run that action automatically or escalate it to a fallback or a human reviewer.

Google's own spill test

In Google's example, the model looks at a photo of a breakroom counter with three questions in one request. It returned has_spill 0.95, next_action wipe_with_cloth at 98%, and a severity score of 1.19 out of 2 with a lower confidence of 0.77. That lower number is exactly the kind of answer you might send to a person.

How to set it up

  1. 1.Get a Gemini API key

    Open Google AI Studio (aistudio.google.com), sign in, and create an API key. Set it as GEMINI_API_KEY in your terminal. The no-cost tier is meant for prototyping.

  2. 2.Install the SDK (version 3.2 or later)

    Python or JavaScript, whichever your app uses.

    pip install -U google-genai
    # or
    npm install @google/genai@latest
  3. 3.Make your first yes/no call

    This is the quickstart from Google's docs. Note the v1beta API version.

    from google import genai
    
    client = genai.Client(http_options={"api_version": "v1beta"})
    
    result = client.decisions.create(
        model="gemini-decision-1",
        input="The blue mug I ordered on Monday arrived cracked on Wednesday.",
        questions={
            "is_damaged": {
                "type": "noul",
                "instructions": "Did the item arrive damaged?",
            },
        },
    )
    
    answer = result.answers["is_damaged"]
    print(f"P(yes)={answer.noul:.2f} confidence={answer.confidence:.2f}")
  4. 4.Add a pick-one question with a way out

    Choice questions always spread probability across your options, even when none fits, so Google recommends planning for that: add a catch-all option, and use the confidence score to send unsure answers to a person.

    result = client.decisions.create(
        model="gemini-decision-1",
        input=ticket_text,
        questions={
            "route": {
                "type": "choice",
                "instructions": "Which team should handle this support ticket?",
                "criteria": {
                    "billing": "Charges, refunds, invoices or plan changes.",
                    "shipping": "Late, lost or damaged deliveries.",
                    "other": "Anything that is not billing or shipping.",
                },
            },
        },
    )
    
    r = result.answers["route"]
    if r.confidence < 0.8:  # pick your own threshold
        send_to_person(ticket_text)
    else:
        route_to(r.choice)
  5. 5.Try Google's demo apps

    Google's launch post links three demo apps in AI Studio: Doorstep Parcels, Speed Sketch and Find the Zoo Keeper. They're a quick way to see the three question types on real images before you write more code.

The fine print

  • Pricing: $0.10 per million input tokens, $0 for output tokens. Each question adds its own instruction and criteria tokens on top of the shared input.
  • On the free tier, Google's pricing page says your data is used to improve Google's products. On the paid tier it isn't. Don't send private customer data while prototyping on the free tier.
  • It only picks from the options you give it. It won't write a reply, an email or a summary. Use a regular Gemini model for that.
  • The speed (150 to 250 ms end to end) and the benchmark results are Google's own numbers. On Google's own chart, TypeSafe Jev 1.14 had a lower total price on the shared evals and led two of the nine accuracy rows. OpenAI also has a Decisions API.
  • Only on the Gemini Developer API (v1beta) for now. No Batch API, context caching, YouTube URLs or gs:// links. Support for Gemini in Google Cloud is listed as coming soon.
  • Treat user input as untrusted, and keep the final say with a person for anything with real consequences (refunds, account actions, safety).

FAQ

What is the Gemini Decisions API?

It's a Google API, powered by the Gemini Decision 1 model, that answers closed questions about your input (yes/no, pick one option, or a rubric score) instead of writing text. Each answer comes with probabilities and a confidence score.

How much does Gemini Decision 1 cost?

$0.10 per million input tokens and $0 for output tokens. There's also a no-cost tier in Google AI Studio for prototyping.

How many questions can I ask in one request?

Up to 1,000 questions per request, all about the same shared input. A choice question can have 2 to 255 options.

What can I send it?

Text, images, audio, video and PDFs, combined in one input of up to 256K tokens. Question instructions themselves are text only.

Can it write answers or replies?

No. It generates zero output text. It only scores the options you define, so pair it with a regular Gemini model when you need written output.

Sources

Facts checked 2026-11-12 against Google's launch post (Nov 11, 2026), the AI Studio Decisions API docs and pricing page, and Google's own benchmark and example figures.

Want help building with tools like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.

Join Shipping Skool

More free guides