Google launches Gemini Decisions API for 1,000 checks

Google made its Gemini Decisions API available through AI Studio on November 11, offering developers a way to evaluate up to 1,000 predefined questions against one shared multimodal submission. The API runs on the Gemini Decision 1 model and is designed to return bounded decisions rather than open-ended prose.

A shared input can contain up to 256,000 tokens of text, images, audio, video or PDFs. Developers can submit yes-or-no checks, multiple-choice questions or requests to score material against a defined rubric. The system returns probabilities and confidence scores, according to Google.

The company says the API is intended for tasks such as choosing tools for AI agents, routing queries to models, checking policy compliance and grading retrieved context. A single video or other multimodal submission can be assessed against multiple questions, allowing developers to reuse the same input across a batch of checks.

Google reports end-to-end latency of 150 to 250 milliseconds and median model-computation time of 50 to 100 milliseconds. It says latency remains nearly flat when a request contains between one and 100 questions. These performance figures come from Google’s product...