Google Releases Gemini Decisions API to Run Fast and Bounded Multimodal Queries

AI · First reported 11 Nov, 18:00 · Updated 18:37

Image: Wikimedia Commons · Ysherazi · CC BY-SA 4.0

Google has launched the Gemini Decisions API powered by its Gemini Decision 1 model, designed to score bounded choices rather than generate free-form text. The API processes multimodal inputs and evaluates up to 1,000 closed-set questions per request at low latency and reduced cost.

Google introduced the Gemini Decisions API and its Gemini Decision 1 model to deliver constrained categorical outcomes rather than generative prose.
The service processes up to 1,000 questions in a single forward pass over shared multimodal inputs, including text, images, video, audio, and PDFs up to a 256,000-token context window.
Inputs support three question structures: yes-or-no assessments, multiple choice selections across 2 to 255 options, and ordered rubric grading scales.
Responses avoid token-by-token decoding, maintaining an end-to-end latency of 150 to 250 milliseconds with model compute times between 50 and 100 milliseconds.
Google prices the service at $0.10 per 1 million input tokens with no charges for output tokens, making it nearly five times cheaper than Gemini 3.5 Flash-Lite across Google's benchmark suite.