AiStaffo

Google launches Gemini Decisions API for 1,000 questions per request

Stock photo: laptop desk documents

Google introduced the Gemini Decisions API on 11 November 2026. It runs on a new model, Gemini Decision 1, and answers up to 1,000 yes/no, multiple-choice or rubric-scoring questions in one request. The shared input can include text, images, audio, video and PDFs. Google lists the price at $0.10 per million input tokens, with no charge for output, and the service is in beta through the Gemini Developer API. For businesses that automate work, the practical point is that software can receive a yes/no or routing answer without reading a generated paragraph. A developer still has to call the API, and the service returns scores rather than text.

In short

  • Google launched the Gemini Decisions API on 11 November 2026, in beta through Google AI Studio and the Gemini Developer API.
  • One request can carry up to 1,000 yes/no, multiple-choice or rubric questions against text, images, audio, video or PDFs.
  • Google lists $0.10 per million input tokens with no output charge; the free tier is limited per minute.
  • The model returns scores and probabilities, not text, so a developer must build the workflow around it.
  • Google Cloud support has been announced but has no release date.

What happened

On 11 November 2026, Google introduced the Gemini Decisions API through the Gemini API in Google AI Studio. According to Google's launch post, the endpoint is powered by a new model called Gemini Decision 1, identified in code as gemini-decision-1. The API evaluates yes/no checks, multiple-choice questions and rubric scores against one shared input, and it can take up to 1,000 questions per request. Google says the input can span text, images, audio, video and PDFs. The service is in beta through the Gemini Developer API.

Why it matters for businesses that automate work

Most generative AI tools answer with prose. Many routine business decisions do not need prose. A bookkeeper needs to know whether a bill has a purchase order number. A claims handler needs to know which queue a file belongs in. According to VentureBeat, Google built the Decisions API for exactly this kind of bounded choice, and its example is a claims intake system that sends a scanned form, damage photos and call notes in one request, then asks which queue should take the claim, whether evidence is missing and how urgently it needs review.

Cost is the other reason the launch matters. Dataconomy reports that Google's pricing example puts a 4,000-token record with 20 checks at about $0.50 per 1,000 records. That figure comes from Google's own worked example, so a real workload will produce its own number.

What changes in practice

For a team that already runs an automation, the main change is how the checks are sent. The shared document is sent once, and the questions travel with it. Google's launch post describes the three question types below. Dataconomy reports the multiple-choice limit as 2 to 255 options.

Question typeWhat comes back
Yes/noA probability that the stated condition holds
Multiple choiceThe winning option and a probability for every option
Rubric scoreA weighted score on a scale the developer defines

Google reports end-to-end latency of 150 to 250 milliseconds for inputs up to 16,000 tokens, staying roughly flat from one to 100 questions per request, as reported by SiliconANGLE. Google's benchmark figures come from Google itself, and independent testing of the model had not been published at the time of writing.

The limits matter as much as the features. The model does not write text, so it cannot draft a reply or explain a decision in words. Dataconomy notes that each question is evaluated independently, which means any dependency between decisions has to be handled in your own code or in follow-up requests. Google Cloud support is coming, but Google has not given a date. Dataconomy also reports that the free tier in Google AI Studio is limited to 200 requests and 50,000 input tokens per minute. A business without a developer will need one to call the API.

What to watch next

  • Whether Google Cloud support arrives, and on what date.
  • Whether the beta price and free-tier limits change before general availability.
  • How the API compares with OpenAI's Decisions API, which InfoWorld reports entered the market weeks earlier.
  • Whether third-party platforms add the model to their own tools.

How AiStaffo would automate this

A typical use is an intake queue for supplier invoices. The AiStaffo system would connect to the mailbox or shared folder where invoices arrive, and send each document once to the Decisions API with a set of checks: is there a purchase order number, does the total match the order, which approval queue should own it, and how complete is the document on a scored rubric. Routine items would be filed and routed automatically, and mismatches would be flagged with the reason. The owner or finance lead still approves payments, decides on disputed items and sets the thresholds the checks use. Results depend on the volume and format of the reader's own documents, which the free audit measures. Book a free automation audit.

Questions people ask

Is the Gemini Decisions API available to use now?
Yes, according to Google's launch announcement, the service is in beta through the Gemini Developer API and is accessible from Google AI Studio. Google says a free tier is available for prototyping. Support for Gemini on Google Cloud has been announced but Google has not set a release date.
How much does the Gemini Decisions API cost?
Google lists the price at $0.10 per million input tokens, with no charge for output tokens. Dataconomy reports Google's example of about $0.50 per 1,000 records for a 4,000-token record with 20 checks. Your actual cost will depend on how much input each request contains.
Does the Gemini Decisions API write answers in text?
No. According to Google's launch material as reported by SiliconANGLE, the model returns terse answers such as yes or no, along with a confidence score. Google describes it as scoring predefined options rather than generating text token by token.
What types of questions can it answer?
The API supports yes/no checks, multiple-choice questions with 2 to 255 options, and rubric scoring on a developer-defined scale. Google says a single request can mix these question types against the same shared input. Each question is evaluated on its own.
Can a business owner use it without a developer?
Not directly. The API is called through code, using the Python, JavaScript or REST guides in Google's documentation. A business without in-house developers would need an automation partner or contractor to build and maintain the connection.

Book a free automation audit

Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

We use these details only to reply to this request and prepare your audit, and keep them for up to two years. We send no marketing unless you tick the box above. Privacy policy

ai modelsai automationdocument processingai agents