Google launches Gemini Decisions API for fast multimodal choices

Google wants its AI to stop writing essays when a simple answer will do.

On November 11, 2026, the company introduced the Gemini Decisions API, a new endpoint built for fast, closed-set decisions. It runs on a new model called Gemini Decision 1, identified in code as gemini-decision-1, and is available through the Gemini API in Google AI Studio.

The pitch is narrow by design. Instead of open-ended generation, the API answers questions that have a fixed menu of possible responses.

What the Decisions API actually does

Google describes the API as a set of purpose-built primitives for bounded choices. The examples it gives are tool selection, policy checks, model routing and context grading.

Developers can feed the API a single multimodal input. That input can span text, images, audio, video and PDFs, with a ceiling of 256K tokens.

Against that one input, developers can ask up to 1,000 independent questions in a single request. So one uploaded video could be checked against a long list of yes-or-no criteria at once, rather than one prompt at a time.

Google frames the outputs as fast and calibrated. Calibration matters here: a calibrated model’s confidence should track how often it is actually right. A model that says it is 90% sure should be correct roughly nine times out of ten.

The company did not disclose pricing alongside the launch.

A crowded autumn for decision models

Google is not the first big lab to ship this kind of product. OpenAI’s Decisions API entered public beta on October 6, 2026. Microsoft followed on October 9, 2026, with a model called Microsoft-Decision-1.

That puts three of the largest AI players into the same niche within roughly five weeks.

Google’s version leans on two selling points: its large input window and native multimodal support.

Why this matters for developers and enterprises

The most immediate audience is teams building AI agents. A dedicated decision endpoint could, in principle, handle micro-choices more cheaply and quickly, leaving heavier generative models for the parts of a workflow that genuinely need them. Without published pricing, though, the cost case remains unproven.

Policy enforcement is another obvious fit. Content moderation and compliance review are, at heart, closed-set questions: allowed or not, compliant or flagged. Running hundreds of policy checks against a single document or video in one request could compress what is now a slow, piecemeal process.

Context grading rounds out the list. Retrieval systems often pull in documents that look relevant but are not. Scoring that material before it reaches the main model can improve answer quality and trim wasted tokens.