HomeInsightsGoogle’s Gemini Decisions API Can Score 1,000 Questions in One Pass
TechnologyNov 12, 2026, 8:44 PM

Google’s Gemini Decisions API Can Score 1,000 Questions in One Pass

Gadriel Borbor Gargard

Published by

Gadriel Borbor Gargard3 min read

Google has introduced Gemini Decisions API, a multimodal service built for bounded choices such as classification, policy checks and model routing. It can evaluate up to 1,000 questions against a shared input in a single pass.

Google Gemini API promotional graphic, used to illustrate the launch of the Gemini Decisions API.

Google Gemini API promotional graphic, used to illustrate the launch of the Gemini Decisions API.

On this page

Google is building AI for decisions, not long answers

Google has introduced the Gemini Decisions API, a new endpoint designed to score choices and return confidence information rather than generate a long response one token at a time.

Announced on November 11, 2026, the service is powered by Gemini Decision 1 and is available through the Gemini API in Google AI Studio.

The idea is straightforward. Many software systems do not need an AI model to write an essay. They need it to choose between options, classify an item, check a policy or decide which tool should handle a task.

Up to 1,000 questions from one shared input

Google says developers can provide one multimodal input and ask up to 1,000 independent questions about it in a single request.

The shared input can contain text, images, audio, video and PDF material, with a context limit of up to 256,000 tokens. The API returns scores, probabilities and confidence information for the candidate choices.

Potential uses include checking whether a document meets a policy, routing a request to the right service, grading context before an AI agent acts, or selecting among a set of possible tools.

This approach could be useful when a system must make many small decisions about the same body of information.

Speed and pricing claims

Google reports end-to-end latency of about 150 to 250 milliseconds. It lists pricing at US$0.10 per one million input tokens, with no output tokens billed because the model scores the available options rather than generating conventional answer text.

Those figures are Google’s published claims. Actual performance will depend on the size and type of input, the questions being asked and the way developers build their applications.

Google also says the model performs strongly on its internal benchmarks for agent routing, reasoning, ranking and multimodal understanding. Independent testing will be useful for judging how well those results translate into everyday applications.

What developers should consider

The Decisions API is aimed at bounded choices. It is not a replacement for every kind of language-model task, and it should not be treated as a source of unrestricted reasoning or open-ended writing.

Developers still need to design clear candidate options, test the model against representative examples and decide what should happen when confidence is low. High-impact decisions should retain suitable human review and safeguards.

For applications that repeatedly classify, route or check information, Google’s new endpoint offers a different way to use multimodal AI. The practical question now is whether its speed and cost hold up under real workloads.

Source: Google: Gemini Decisions API

Similar insights