Google announced the Gemini Decisions API on November 11. It runs on a new model, Gemini Decision 1. A developer sends one input, up to 256K tokens, which can mix text, images, audio, video and PDFs. Up to 1,000 questions can be asked about it: yes/no, pick from 2 to 255 options, or a rubric score. Each answer has a confidence score from 0 to 1. Tech in Asia also covered the launch.
Google says the model rates the options directly rather than writing an answer word by word. It puts median delay at 150 to 250 milliseconds, start to finish, for inputs up to 16K tokens, and says speed stays nearly flat from 1 to 100 questions. Price is $0.10 per 1M input tokens, with no output charge. On Google's own benchmarks, it says the cost is nearly 5x lower than Gemini 3.5 Flash-Lite with the Interactions API, and that it has the lowest calibration error, meaning confidence scores that match real outcomes. No outside tests are cited. Google Cloud support has no date.
Routing, policy checks and grading are fixed-choice jobs. Because this model writes no answer text, Google says there are no output or thinking-token bills. Buyers can test how often the confidence scores are right on their own data, and set the score below which a person reviews the case.