Skip to content
D Consultants

Gemini Decisions API is live: how UAE businesses can cut the cost of AI agent decisions

Google launched the Gemini Decisions API on November 11, 2026. It answers yes/no, multiple-choice and rating questions with a confidence score in a fraction of a second. Here is where it fits in your AI agents and how to test it on one workflow.

Published 4 min readD Consultants team
Glowing data tiles flow through a geometric gate and split into three paths, one highlighted, with Dubai's skyline beyond
Share this article
Summarize with AI

Key points

  1. 01Google launched the Gemini Decisions API and the Gemini Decision 1 model on November 11, 2026.
  2. 02One request can carry up to 1,000 yes/no, multiple-choice or rubric questions against one shared input.
  3. 03Google lists $0.10 per million input tokens, no output charge and 150-250 ms end-to-end latency.
  4. 04Every answer returns a confidence score, so unclear cases can go to a person.
  5. 05Start with one high-volume decision step and test it against your current model.

Google launched the Gemini Decisions API on November 11, 2026, powered by a new model called Gemini Decision 1 (Google). It is built for one job: picking from a fixed set of answers quickly and cheaply, instead of writing free text.

What changed

Many AI agent calls are small decisions. Is this lead qualified? Which team owns this ticket? Until now, those checks usually ran on a general model that writes a full answer, token by token.

The Decisions API takes a different route. According to Google's announcement, it supports three question types: yes/no, multiple choice with 2 to 255 options, and scores on an ordered rubric. One request can pair a single input of up to 256K tokens, across text, images, audio, video and PDFs, with up to 1,000 independent questions. Every answer comes back with a confidence score between 0 and 1.

Google lists the price at $0.10 per million input tokens with no charge for output tokens, and reports 150 to 250 milliseconds end to end (Google). Google says this setup is nearly five times cheaper than Gemini 3.5 Flash-Lite across its own benchmark suite. Test that on your own data.

Why it matters for UAE businesses

The cost of an AI agent rarely sits in the main answer. It sits in the many small checks around it. A property enquiry gets classified, scored and routed before anyone replies. A support ticket gets tagged by language, urgency and product. A voice call gets reviewed after it ends. These steps run on every interaction, so their unit cost decides whether an agent pays for itself.

Three details matter for teams here. First, the input is multimodal. Google's own examples include pairing a call recording with account data, or a scanned claim PDF with damage photos (Google). Second, the confidence score gives you a clear rule for sending unclear cases to a person. Third, the API runs on the v1beta Gemini API in Google AI Studio, and Google lists support in Gemini in Google Cloud as coming soon (Google). If your data has residency or contract limits, confirm where it is processed before you send customer records.

This does not replace your main model. It handles the routing and checking around it. In AI agents and automation projects, that layer is the first place we review when an agent costs more than planned. The same logic applies to business automation and CRM workflows, such as those in atnos, where rules sort and score every record.

What to do now

Your next steps

  1. List your closed decisions

    List every step in your agents and automations that ends in yes/no, a pick from a list, or a score. Note how often each one runs.

  2. Pick one high-volume step

    Choose the step with the most volume and the clearest right answer, such as lead qualification or ticket routing. Build a test set of real, anonymized examples with known answers.

  3. Compare against your current model

    Run the same test set through your current setup and through the Decisions API, using the no-cost tier in Google AI Studio. Compare accuracy, latency and cost per thousand decisions.

  4. Set a confidence threshold

    Decide the score below which a case goes to a person or a stronger model. Log every handover so you can tune the threshold over time.

Questions

Does the Decisions API replace our chatbot or voice agent model?

No. It answers closed questions with fixed options. Your agent still needs a generative model to talk to customers.

Is it cheaper in every case?

Google reports that it is nearly five times cheaper than Gemini 3.5 Flash-Lite on its own benchmark suite. Your result depends on input size and how many questions you batch per request, so measure on your own data before you switch.

Can we use it with Arabic content?

Google's announcement does not list supported languages. Include Arabic and mixed Arabic-English examples in your test set and check accuracy before you rely on it.

Tags:GeminiAI agentsGoogle AIAutomationAI costs
Share this article

Written by

D Consultants team

Software and AI team, Dubai

The people who build and run dconsults.com, our own products atnos, Vline and blaaa, and websites and AI agents for businesses across the UAE.

Profile

More articles to explore

All articles

Tell us what you are building.

Free 30-minute call. We map your workflow, find the highest-impact opportunity and send a clear proposal.

+971 50 807 0826