Google launches Gemini Decision 1, a model that answers up to 1,000 yes/no or multiple-choice questions in one pass

Google DeepMind has released the Gemini Decisions API, built for quick, closed-ended calls instead of open-ended chat, Matthias Bastian reports for The Decoder.

What it does

The model behind it, Gemini Decision 1, handles three kinds of questions: yes or no, pick one of several options, and rate on a scale. It can take up to 1,000 separate questions at once, over text, images, audio, video and PDFs. Instead of writing an answer, it rates each option and hands back probabilities plus a confidence figure, typically within 150 to 250 ms.

Google's demo puts all three question types to a single kitchen photo: is there a stain, which cleaning step fits, and how bad is it on a scale.

Where Google already uses it

Internally, Google says it routes requests between models in Gemini Notebook, helps with design prototyping in Stitch, and runs safety checks in Search.

Price and availability

It costs $0.10 per million input tokens and nothing for output, roughly a fifth of what Gemini 3.5 Flash-Lite costs, by Google's count. It's in Google AI Studio now, with OpenRouter, LangChain, Vercel and others coming. Google's own benchmarks put it first in categories such as agent routing, reasoning and multimodal understanding, with the best calibration of the group.

A crowded new category

Decision models are suddenly everywhere. Startup Jev shipped one first, and OpenAI, Cloudflare and Microsoft followed quickly.

Scoreboard update: AI +1. A model now makes a thousand small calls in a quarter of a second.

Stay undead.