Google Releases Gemini Decisions API Powered by Gemini Decision 1 Model

Google has rolled out the Gemini Decisions API, utilizing a new model named Gemini Decision 1. Unlike traditional language models, this system does not produce text outputs. Instead, it evaluates up to 1,000 typed questions against a single shared input to deliver calibrated probabilities. According to the company, the tool is aimed at streamlining minor branching tasks inside autonomous agents, including policy checks, routing, tool selection, and grading.

The system supports multimodal data, allowing developers to process text, images, video, audio, and PDF files within a single prefill pass that spans a 256K token context window. Rather than generating text token by token, the model scores candidate options simultaneously in a single forward pass, achieving response times between 150 and 250 milliseconds end-to-end. Google prices the service at $0.10 per one million input tokens, with no charges applied for output tokens.

Developers can structure requests using three distinct question categories: yes/no queries returning a probability boundary, multiple-choice options between 2 and 255 items, and rubric-based scoring levels. Each returned answer includes a confidence metric ranging from zero to one. The API is hosted exclusively on the Google Developer API platform without open weights or self-hosting options.

What to watch next: Monitor developer adoption rates and whether competitors introduce similar structured decision-making APIs.