Google launches Gemini Decisions API, decision-only endpoint

Google's move follows similar efforts by Jev with TypeSafe, Microsoft with Decision-1, and OpenAI with a Decisions API. What Google is emphasizing as different is the ability to include multiple materials in a single input — multimodal support for text, images, audio, video, and PDFs up to 256,000 tokens — and to run large volumes of checks at relatively low cost.

The design reflects a trade-off: by not generating text token by token, the API returns only answers and confidence scores, which Google says saves processing cost and speeds up responses. It also allows mixing Yes/No, choice, and score question types within one input, and flagging low-confidence answers for human operators or another inference model.

API keys are available through Google AI Studio, giving developers a direct way to test the decision-focused endpoint against existing classification and moderation workflows.