Google on Wednesday introduced the Gemini Decisions API, a new endpoint powered by the company’s Gemini Decision 1 model that evaluates questions across multiple media formats.
The company said the service helps developers make probabilistic choices to route queries, grade content, or check policy rules. Rather than generating text word by word, the model scores candidate options in a single forward pass and returns confidence scores between zero and one, according to Google. The company stated that the new model led a suite of benchmarks for agent routing, reasoning, and multimodal understanding.
The API can process up to 1,000 yes-or-no, multiple-choice, and rubric-scoring questions in one request, the company said. It accepts combinations of text, images, audio, video, and PDF files within a single shared input of up to 256,000 tokens. Google estimates the system returns responses with an end-to-end latency of 150 to 250 milliseconds.
The system allows developers to pair different file types in one input, such as combining a call recording with account metadata or a scanned claim with damage photos, the company said. The API handles yes-or-no questions by estimating the probability of a condition, evaluates multiple-choice queries by returning a winner from up to 255 options, and assesses rubric scores by providing a continuous weighted grade.
The service costs $0.10 per 1 million input tokens, with no charges for output tokens. Google said this pricing makes the tool nearly five times cheaper across a benchmark suite than its Gemini 3.5 Flash-Lite model. Shared inputs are billed once per request, and the company noted there are no separate charges for audio transcription or optical character recognition.
Several Google teams are already using the tool, including for dynamic model routing in Gemini Notebook and safety checks in Google Search, the company said. The API is available now through Google AI Studio. Google stated that support for Gemini in Google Cloud is coming soon.