Google's Gemini Decisions API bills $0 for output

Stop paying a chat model to answer yes or no.

Google launched the Gemini Decisions API on Nov. 11, built on a model called Gemini Decision 1. It scores answers instead of writing them.

One input, up to 1,000 questions, no output bill

You send one shared input and a list of questions about it. Google says the model handles up to 1,000 independent questions per request, on inputs up to 256K tokens, in text, images, audio, video or PDFs.

There are three question types, per Google's post: yes/no, multiple choice with 2 to 255 options, and an ordered rubric that returns a weighted score. Each answer comes back with probabilities and a confidence score from 0 to 1.

The model scores the options in a single pass instead of generating text token by token, Google says. It quotes 150 to 250 milliseconds end to end, with latency nearly flat from 1 to 100 questions per request.

The price:

$0.10 per million input tokens. $0 for output.

The shared input is billed once per request, plus each question's own instruction tokens, according to the post. Google says that makes it nearly five times cheaper than Gemini 3.5 Flash-Lite across its benchmark suite. That comparison, like the benchmark lead Google claims, is Google's own.

Routing is now a product

Most apps that run on AI quietly spend tokens on small decisions. Which model should answer this? Is this message safe? Does this draft pass the rubric? Today you usually ask a chat model and parse the text it writes back.

Google is betting those calls deserve their own tool. It says it already uses Decision 1 inside its own products: to route queries in Gemini Notebook, where it reports up to 40% lower latency for everyday queries, and as a safety layer in Google Search that detects adversarial behavior in MCP tools.

Here's the thing. The confidence score matters more than the price. A probability you can trust lets your code send the unsure cases to a bigger model or a person, and wave the rest through. That is model calibration turned into a feature, and Google says Decision 1 posts the lowest calibration error on its suite.

Where to try it

It's in the Gemini API through Google AI Studio now, with a no-cost tier for prototyping. Google lists integrations rolling out with OpenRouter, LangChain, Vercel, LiteLLM and Pydantic AI, among others. Support in Google Cloud is "coming soon," the post says.

Good first jobs for it:

• Model routing: pick the model tier and reasoning level for each request.
• Moderation and policy checks on user content.
• Grading outputs against a rubric in your evals.
• Tagging images, calls or documents against a fixed list of labels.

Open-ended questions still need a chat model. Decision 1 sits in front of one.

If you already use Gemini 4 Carbon or another frontier model for these jobs, the math is simple. Take a week of your routing or moderation calls, replay them through Decision 1, and compare accuracy, latency and cost.

Watch the calibration hold up outside Google's lab

Google's numbers come from Google's suite. The test that counts is yours: does a 0.9 confidence mean right nine times in ten on your data?

Then watch Google Cloud availability. Enterprise teams that can't use AI Studio are waiting on it.

Run your routing calls through it. Keep whichever is cheaper and still right.