Google has launched Gemini Decision 1, a multimodal decision model designed specifically for closed-ended multiple-choice questions, along with the accompanying Decisions API. The model features ultra-low latency and native multimodality, can process 1,000 questions at once, and supports three native decision types: Yes/No judgments, multiple choice with up to 255 options, and scoring against criteria. In terms of pricing, each million input tokens costs only $0.1, output tokens are free, and the cost is only one-fifth that of Gemini 3.5 Flash-Lite. Technically, the model relies on Gemini's native multimodal architecture, can process up to 256K tokens of mixed multimodal data including text, images, audio, video, and PDFs at once, and completes scoring of all options through a single forward pass. It performs excellently in benchmark tests, with extremely low calibration error and a low hallucination rate. The median computational latency for processing a 16K-token multimodal context is only 50 to 100 milliseconds, and the three decision modes can be mixed within a single request while returning confidence scores from 0 to 1. In terms of applications, Google has already used it internally as a low-latency control hub, deploying it in scenarios such as dynamic model routing, rapid design iteration of product workflows, quick emoji matching, and search safety detection. Google AI Studio provides a free API tier for prototype development, and a large number of mainstream AI development platforms and tools are advancing related integrations. Recently, the decision model track has seen fierce competition among giants, with multiple vendors launching similar products in succession within 17 days, and Google is entering with a differentiated approach of native multimodality plus large-scale concurrency for thousands of questions. The industry has gradually formed a new Agent architecture in which decision models handle ultra-fast, low-cost judgments while general large models handle deep, slow thinking.