Google launches Gemini Decisions API: faster, cheaper AI for yes/no, multiple choice & rubric scoring

Google has launched Gemini Decisions API, featuring a new model called Gemini Decision 1, designed specifically for high-speed decision-making tasks. Unlike traditional language models that generate text token by token, this system scores options directly, which dramatically reduces latency to 50–100 milliseconds of model compute (150–250 ms end-to-end), and keeps costs roughly five times lower than using Gemini 3.5 Flash-Lite for similar tasks. It handles multimodal inputs—text, images, audio, video, and PDFs—in a single forward pass, supports contexts up to 16K tokens, and can evaluate up to 100 yes/no, multiple choice, or rubric-scoring questions per request.

This API is aimed at developers building automated review systems—like processing insurance claims with photos, grading assignments, or routing customer support tickets—where speed, cost, and calibrated confidence scores matter. Each decision comes with a confidence score from 0 to 1, letting applications flag low-confidence cases for human review. The takeaway: Google is pushing beyond pure text generation into fast, cheap, multimodal decision engines optimized for production workflows.