Google launches Gemini Decision 1, an AI that returns probabilities instead of text, at $0.10 per 1M tokens with zero output tokens

Google has added Gemini Decision 1 to the Gemini API, a new model that skips writing text and returns calibrated judgments and probabilities for the questions it is asked. It costs $0.10 per 1 million input tokens with no charge for output, and developers can try it free in the API.

According to an X post on Nov 11 by Logan Kilpatrick, product lead for Google AI Studio, and the new Decisions API documentation, the model reads text, images, audio, video and PDFs natively and takes up to 1,000 questions in a single request. Answers come in only three forms: a yes/no probability, one option from a fixed list, or a score on a rubric, each with a confidence value. The model "generates zero text tokens, so output token usage is always 0," the docs say, so code can branch and sort on the numbers directly instead of parsing prose. The context window is 256K tokens, and the docs' examples include judging package damage from a photo and routing support tickets. In Google's comparison chart, it topped 7 of 9 benchmarks against OpenAI's Decisions API running GPT-6 Luna and TypeSafe's Jev 1.14, though running seven shared evals cost $0.88, more than Jev's $0.68.

"Gemini Decision started organically inside Google DeepMind," said Google DeepMind's Philipp Schmid, adding that teams across Google picked up an internal prototype fast and "6 weeks later we're shipping a new model and a new API together." Kilpatrick's post passed 2,500 likes within an hour. Replies ranged from "new SOTA" to a warning that "the free tier keeps your data for training."