Google has released the Gemini Decisions API, a developer tool that scores a fixed set of options and returns probabilities instead of written text. It is an API for people building software, not an app you can open on your phone, according to Google's announcement.

What it does

Most AI models answer by writing text one piece at a time. Google says this endpoint, powered by a new model called Gemini Decision 1 (model id gemini-decision-1), instead scores your candidate options in a single pass. It returns probabilities and confidence scores between 0 and 1. Google calls the target "fast, closed-set decisions," meaning questions with a fixed list of possible answers.

There are three question types:

  • Yes/no: the probability that a condition holds.
  • Choice: pick from 2 to 255 options, and get the winner plus normalized probabilities.
  • Rubric score: rate against an ordered rubric starting at 0, and get a weighted score plus the probability of each level.

Google says a developer's code can use the confidence score to send ambiguous cases to a fallback or a human reviewer.

What you can feed it

Google says one request can combine text, images, audio, video and PDFs, up to 256K tokens, with up to 1,000 independent questions per request. A token is a small chunk of text or data that AI models count and bill by.

Google's figures for speed are 150 to 250 milliseconds end to end, of which 50 to 100 milliseconds is model compute. We have not measured that.

One example from Google

Google's post uses a breakroom counter. A developer sends a photo of the counter plus a text note in one request, and asks three questions: is there an uncleaned spill, which cleanup action fits, and how severe is it. The answers come back as probabilities, not a paragraph.

What it costs

Per Google, input is $0.10 per 1 million tokens, and output tokens are billed at $0. The Gemini API has a no-cost tier. Google says you are billed once per request regardless of how many questions you ask, and each extra question adds only the tokens for its own instruction or criteria. It lists no thinking-token charge and no separate transcription or OCR charge.

Google also says the model is "nearly 5x cheaper" than Gemini 3.5 Flash-Lite with the Interactions API across its shared benchmark suite. That is Google's own benchmark claim, and we have not verified it.

How to try it

Google says developers can start today with an API key from Google AI Studio. Its docs cover Python, JavaScript and REST. For other recent Gemini developer news, see our reports on Gemini 4 Carbon in the API and Gemini 4 Argon in the Gemini app.

What we checked

We read Google's announcement and the Gemini API release notes entry dated Nov. 11. We have not tested the API, and we found no independent benchmarks yet. The speed, price comparison and accuracy figures are Google's.

What we don't know

We have not read details on Gemini Decision 1's size or architecture, any independent accuracy results, or the API's rate limits. For more, see AI apps.