Google has launched the Gemini Decisions API, an endpoint built for fast closed-choice calls, along with the model behind it, Gemini Decision 1. It arrived in Google AI Studio on 11 November, and Vercel's AI Gateway, where it appears as google/gemini-decision-1, and OpenRouter listed the model the same day.
“Teams across Google are already using the Decisions API as a low-latency control plane, shipping features simply by defining options rather than training custom classifiers or waiting on generative calls”
Google, in its announcement post · blog.google ↗
Instead of writing text token by token, the model scores the options in a single forward pass. One request takes an input of up to 256,000 tokens and as many as 1,000 independent questions. Questions can be yes/no, multiple choice with 2 to 255 options, or a score on an ordered rubric, and every answer comes back with a confidence between 0 and 1, so code can send shaky cases to a fallback or a person.
Google puts end-to-end latency at 150 to 250 milliseconds, barely changing from one question to a hundred. Input costs $0.10 per million tokens, output costs nothing, and AI Studio has a free tier for trying it. Google says that is about five times cheaper than Gemini 3.5 Flash-Lite on its own benchmark suite, and that the model leads on agent routing, reasoning, ranking and multimodal understanding with the lowest calibration error.
Google already uses it inside its own products. Gemini Notebook routes requests between models with it, cutting latency by up to 40%, and it also drives Stitch's rapid canvas, emoji reactions in a Google Labs tool and a safety layer for Search's MCP tools. Integrations are rolling out across OpenRouter, Vercel, LangChain, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, with Gemini in Google Cloud listed as coming soon and more integrations still to follow. The endpoint reuses the content-block input format of the Interactions API that serves Gemini 3.5 Flash-Lite, and developers call it through client.decisions.create(), and Google also built three demo apps: Doorstep Parcels, Speed Sketch and Find the Zoo Keeper.
About Gemini Decision 1
Gemini Decision 1 is Google's model for its Decisions API. It answers structured questions about an input, either true/false, multiple choice with up to 255 options, or a score with up to 255 levels, and it returns a confidence for every answer.
It gives typed decisions only, with no free-form prose or reasoning trace, and Google has not disclosed its parameters.
Nov 2026
Released
AlphaSignal, which wrote up the launch, adds a caution: the benchmark figures are company-reported, the endpoint returns no prose or reasoning trace so open-ended work still needs a generative model, and Google has not said how big the model is. Outlets also differ on input types and context size: Google lists text, images, audio, video and PDFs at 256K, OpenRouter shows 262K, and Vercel's version takes text and images only.
Sources · 3