adtestbench Higgsfield
EN

AI news · Google · Gemini Decision 1

Google’s Gemini Decision 1 scores up to 1,000 questions at once

The new model answers yes-or-no, multiple-choice and rubric questions about text, images, video or PDFs in a quarter of a second, with no charge for its answers.

By Alexandre S. , 06:30 UTC

Google released the Gemini Decisions API on Wednesday 11 November 2026, built on a new model, Gemini Decision 1. It does one job: it takes a single input, such as a video, a set of images, a PDF or text of up to 256,000 tokens (the word fragments AI models count by), and answers up to 1,000 closed questions about it, Google says. Each question is a yes or no, a pick from up to 255 options, or a score on a scale you define, and every answer carries a confidence score from 0 to 1.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a microchip standing on the horizon.
Drawn by adtestbench from “Introducing the Gemini Decisions API”,

Gemini Decision 1 scores every possible answer in one pass, with no text to write out, so a reply takes 150 to 250 milliseconds, Google says. Input costs $0.10 per million tokens, answers cost nothing, and one request pays for its input once whatever the number of questions. Google lists policy checks and content grading among the uses.

Google is late to this category. Microsoft, Cloudflare and OpenAI all launched rivals after Jev, The Decoder notes; Jev was the first such model, released in mid-September by the start-up TypeSafe, and Google says its own model beat Jev and OpenAI’s version on seven of nine benchmarks while Jev came out slightly cheaper, SiliconANGLE reports.

The text rates on our price sheet, checked on 11 November, show the gap: OpenAI charges $5 per million text input tokens to GPT Image 2.6, fifty times what Decision 1 charges to read the same text.

Since the input is paid once per request, checking an ad video against 100 brand rules costs about the same as checking it against one: the bill follows the video’s length, and each extra rule adds only its own few tokens.