On November 11, 2026 Google introduced the Gemini Decisions API, a service that does not write text: it makes a decision. You give it some content (a text, an image, a PDF, an audio file or a video) and a set of closed questions, and it answers with a yes or no, a choice between options or a score, always with a confidence level.
It sounds like a developer detail, but it is the kind of building block that makes automations work: sorting emails, checking that a document follows a rule, deciding which tool to use. The figures below are the ones Google states in its announcement and have not been verified by third parties.
What is the Gemini Decisions API?
It is an interface (API) built on a dedicated model that Google calls gemini-decision-1. Unlike a chatbot, which generates an answer word by word, this model scores the options in a single pass. That is why, according to Google, it responds in about 150-250 milliseconds.
Every answer includes a confidence score from 0 to 1. If the score is low, you can send the case to a manual check or to a more powerful model instead of trusting it blindly.
What can it do in practice?
| Question type | Example |
|---|---|
| Yes or no | Does this contract contain an automatic renewal clause? |
| Multiple choice (2 to 255 options) | Which department should this request go to? |
| Score with a rubric | How clear is this text, from 1 to 5? |
- A single input, up to 256,000 tokens, can be evaluated with up to 1,000 questions in one request.
- It accepts text, images, audio, video and PDFs.
- Google lists tool selection, policy checks, model routing and content grading as uses. According to the company, Gemini Notebook uses it to route queries and Google Search uses it as a safety layer for MCP tools.
How much does it cost?
- $0.10 per million input tokens, with no charge for output tokens.
- The shared input is billed once per request, however many questions you ask.
- There is a free tier in Google AI Studio for testing.
- Google states a cost almost 5 times lower than Gemini 3.5 Flash-Lite with the Interactions API, on its own test suite.
These are comparisons chosen by Google: before building a project on it, try the model on your own documents.
What does it change for people who work?
If you do not code, you will not use the API directly, but you will see it inside the tools you use: faster filters, automatic routing of requests and compliance checks on documents. If you work with automations (for example with no-code tools), the news is that a check like "is this file an invoice?" can cost a fraction of a cent and come back in a fraction of a second. This assessment is ours, based on the data in the announcement. To choose the right model for more general tasks, also read Gemini 4 Hydrogen and Carbon.
What is unclear?
- Test results and the cost comparison come from Google, with no independent verification.
- Google writes that support in Gemini on Google Cloud is "coming soon", and integrations with outside services (such as OpenRouter, LangChain, Vercel, LiteLLM) are described as rolling out.
Frequently asked questions
Does the Decisions API replace ChatGPT or Gemini?
No. It does not write text or hold a conversation: it makes quick decisions on closed questions, usually inside an automated process.
Can I try it for free?
According to Google there is a free tier in Google AI Studio, meant for prototypes. For continued use you pay $0.10 per million input tokens.
What is the confidence score for?
It shows how sure the model is of its answer. When it is low, you can have a person or a more powerful model check the case.
Official sources
This article is written with the help of artificial intelligence models, starting from the official sources listed, and is reviewed before publication.














