Let AI make the quick decisions for people on "right or wrong, which one to choose," and Google has pushed latency down to a matter of milliseconds. On November 11, Google announced the launch of the Gemini Decisions API, backed by the all-new Gemini Decision 1 model, specifically designed for scenarios that require repeatedly making yes-or-no choices, selection decisions, and scoring.
Feed in a thousand questions at once, get scores directly from a single forward pass
Developers upload a multimodal set of materials consisting of text, images, audio, video, or PDFs (up to 256K Tokens), and can throw in as many as 1,000 unrelated questions. Gemini Decision 1 runs just one forward pass and assigns probability and confidence scores to each option—no need to split it into hundreds of calls like the traditional process. This kind of work is perfect for tool selection, strategy verification, model routing, and context quality assessment: for example, feeding in a call recording together with account information, or pairing a scanned claim form with damage photos, and letting the model verify it all in one go.
Fast and cheap, with adjustable resolution
In terms of speed, end-to-end processing latency falls between 150 and 250 milliseconds; pricing is also restrained, at only $0.10 per million input Tokens (about 0.67 yuan), while output Tokens are completely free. Developers can also manually adjust the resolution of images and videos to find a balance between visual detail and Token cost. At present, the API has already launched on Google AI Studio, offering a free quota for prototype development.
