Google makes a Decisions API with a result in 150 to 250 ms

Google made the Gemini Decisions API in Google AI Studio. It uses a new model, Gemini Decision 1. The model finds the correct alternative in a set of alternatives, for example which tool or which model to use. It does not write text. It gives a score for each alternative and a confidence value. Google tells that the total time is 150 to 250 milliseconds.

How it works

You send one input of text, images, audio, video or pdfs. You add a maximum of 1,000 questions. A question is a yes or no check, a selection from alternatives, or a score on a rubric. Each result has a confidence score from 0 to 1. Your code can send a result that is not clear to a person.

The numbers

Google tells that the model is first on its tests for routing, reasoning and ranking, with the minimum calibration error. Calibration shows if a confidence score is the same as how frequently the result is correct. Gemini Notebook uses it to send each query to a model. Google tells that the time is a maximum of 40% lower.

What is not known

Google gave no scores. The cost figure is from a test set of Google. No external test is given. Gemini in Google Cloud does not have it at this time.

Insight: Two large labs have a fast decision API at one price

OpenAI opened its Decisions API in October 2026 for $0.10 for 1M input tokens. OpenAI tells that it competes with a start-up with a value of $7.6 billion. Google has the same price at this time. This reads as a race to control the small, fast check that sends each task to a model. A start-up with only this check has two large labs against it. Test the two APIs before you make a router.