Google launches Gemini Decisions API powered by Gemini Decision 1 at 0.10 dollars per million input tokens with free output
Google on November 11 launched the Gemini Decisions API and the Gemini Decision 1 model, which scores up to 1000 questions over text, image, audio, video and PDF content in a single forward pass and returns probabilities and confidence, at 0.10 dollars per million input tokens with no output charge.
Google announced on November 11 the launch of the Gemini Decisions API for fast multimodal decision making, driven by the new Gemini Decision 1 model. In a single fast pass it can score up to 1000 yes-no judgements, multiple-choice selections and rubric-based evaluations over text, image, video and PDF content.
The API targets decision tasks such as tool selection, policy checks, model routing and context quality assessment. Developers can supply one multimodal input containing text, image, audio, video and PDF, up to 256K tokens, and ask up to 1000 mutually independent questions about it. Gemini Decision 1 scores candidate options through a single forward computation and returns probabilities and confidence scores. End-to-end latency runs 150 to 250 milliseconds, pricing is 0.10 dollars per million input tokens, and output tokens are not charged.
The API is built on Gemini's native multimodal architecture, so developers can combine modalities within a single input, for example pairing a call recording with account metadata or a scanned PDF claims form with damage photos, and can adjust image and video input resolution to control visual detail and token cost. It is available in Google AI Studio with free quota for prototyping. Google chief technology officer Sundar Pichai said on social media that the model is designed for decision tasks over closed option sets and can score up to 1000 questions in one pass. Unlike models that generate text token by token, it returns judgements, probabilities and confidence directly rather than requiring a generative model to write out an answer that then has to be parsed.