Commit

google launched the decisions api, with a new model, gemini decision 1. instead of generating text, it just picks from the options you define. it can be yes or no, multiple choice or a score on a scale, and every answer comes with a confidence from 0 to 1. you can send up to 1,000 questions at once about the same input, which can be text, image, audio, video or pdf. it answers in 150 to 250 ms and costs $0.10 per million input tokens, with no charge for output. google itself already uses it in gemini notebook to decide which model answers each question. a lot of what we do today by calling a whole model just to get a "yes" or a "no" fits here. classifying leads, picking which tool an agent uses, checking if an image has what it should. haven't tested it yet, but it looks really useful. https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/…