Google Ships Gemini Decisions API for Fast Yes-or-No AI Calls

Google introduced the Gemini Decisions API on November 11, 2026, an endpoint aimed at questions with a fixed menu of answers rather than open-ended generation. It runs on a new model called Gemini Decision 1, tagged gemini-decision-1 in code, and ships through the Gemini API inside Google AI Studio. The company pitches it as a set of purpose-built primitives for bounded choices.

Developers hand it one multimodal input — text, images, audio, video or PDFs — capped at 256K tokens. Against that single input they can fire up to 1,000 independent questions in one request, so a lone video can be tested against a long checklist of yes-or-no criteria at once instead of one prompt at a time. Google describes the outputs as fast and calibrated, meaning a model that says it is 90% sure should be right about nine times out of ten.

Google names four uses: tool selection, policy checks, model routing and context grading. Agent builders are the most immediate audience, since a dedicated decision endpoint could handle micro-choices cheaply and leave heavier generative models for work that truly needs them. Content moderation and compliance review are also natural fits, because allowed-or-not and compliant-or-flagged are closed-set calls. Context grading targets retrieval systems that drag in documents which look relevant but are not.

Google kept pricing under wraps at launch. It also arrived late to its own party: OpenAI's Decisions API hit public beta on October 6, 2026, and Microsoft followed on October 9, 2026, with Microsoft-Decision-1. Three of the biggest AI players piled into the same niche inside roughly five weeks, and Google is leaning on two edges — a large input window and native multimodal support.