Google's Gemini Decisions API: AI that decides in 150 milliseconds

Google launched the Gemini Decisions API on 11 November 2026, powered by a new model called Gemini Decision 1. It does one thing, and it does it fast: it answers up to 1,000 yes or no, multiple choice, or scored questions about a piece of text, an image, audio, a video, or a PDF, in a single request. The whole round trip takes 150 to 250 milliseconds, and Google charges $0.10 per million input tokens, with no charge for output tokens, because the model scores options instead of writing an answer word by word.

That sounds like a developer tool, and it is one. But the idea behind it matters to any small business with a website: some questions do not need an essay. They need a decision, with a confidence score, so your software can act on it or send it to a person.

What the Decisions API actually does

Most AI tools you know generate text. You ask a question, the model writes an answer one token at a time, and you wait. The Decisions API works differently. You send one input, for example a photo of a damaged product together with the customer's message, plus up to 1,000 questions about it. Gemini Decision 1 scores every option in a single forward pass and returns probabilities and a confidence score from 0 to 1 for each answer.

The input can be text, images, audio, video, or a PDF, up to 256,000 tokens. Every answer carries a confidence score, so your code can route clear cases automatically and send uncertain ones to a human reviewer. Google reports median model compute of 50 to 100 milliseconds, with 150 to 250 milliseconds end to end, and says the cost stays nearly flat from 1 to 100 questions per request, because the shared input is billed once.

The three question types

There are three kinds of questions, and they map neatly onto business decisions.

First, yes or no questions. Is this enquiry a good fit for our service? Does this photo show damage? Does this review need a reply today? The model returns the probability that the condition holds.

Second, multiple choice. Pick from 2 to 255 options, and get the winner plus a probability for every candidate. Which product in our range fits this customer best? Which team should handle this ticket?

Third, scored rubrics. Grade an input on an ordered scale, for example from clean to major hazard, or from cold lead to hot lead, and get a continuous score with per level probabilities.

What it costs

The price is $0.10 per million input tokens and $0 for output tokens. Google says the Decisions API is nearly five times cheaper across its shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API. There is a no cost tier in Google AI Studio for prototyping. For a small business, the honest way to read the price is per completed job: a lead scored, a ticket routed, a claim checked. At this price, scoring a thousand enquiries costs cents, not euros.

Four small business uses

Lead qualification. A contact form arrives with a message and a budget range. Instead of a person reading every enquiry, the API scores fit, urgency, and budget, and only the good ones reach your inbox. Low confidence scores go to a human, so no real customer is lost to a wrong score.

A product finder. A customer uploads a photo or describes a problem. The API picks the best product from your range, or the right service package, and your website shows the recommendation with a short reason.

Support triage. An email arrives with a photo of a broken part. The API checks whether it is a warranty case, how urgent it is, and which team should handle it, before a person ever opens the ticket.

Content checks. A review, a comment, or a supplier document is scored against your own rules: does it need a reply, does it break a policy, is it spam? The checks run in milliseconds, so they can sit inside a form instead of a nightly batch job.

How to start

Start with one decision, not ten. Pick the question your team answers most often by hand, and write it as a yes or no question with clear criteria. Then build a small test: run 50 real examples through the API, compare the scores with what your team decided, and check where the model was wrong. Keep a human in the loop for every score below your confidence threshold, and review the mistakes weekly for the first month. That test costs almost nothing, and it tells you whether the tool fits your business before you change any workflow.

If you want help adding one scored decision to your website, see our workflow automation service or our AI chatbot service.

What to watch

A confidence score is not a guarantee. It tells you how sure the model is, not whether it is right. Set a threshold, send everything below it to a person, and log the cases where the model was confident and wrong. Those cases are your training data for better criteria. Also remember that the API reads the input you send it, so do not send personal data you would not put in a form. Keep the input minimal, and check your privacy notice before you connect a customer form to any AI service.

The takeaway

The Decisions API is not another chatbot. It is AI that answers with a decision and a number, fast enough to sit inside a form, a checkout, or a support inbox. For a small business, the win is not the technology. It is the hour a day your team stops spending on sorting, and starts spending on customers.

Source: Google, "Gemini Decisions API: Fast multimodal decisions", 11 November 2026, https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/