Overview
- Google announced the Gemini Decisions API on Nov. 11, 2026, introducing the gemini-decision-1 model as a purpose-built endpoint for bounded yes/no, multiple-choice and rubric scoring.
- The model accepts one shared multimodal input—text, images, audio, video or PDFs—supports up to 256K tokens, and can evaluate as many as 1,000 independent questions in a single request.
- Instead of generating text token by token, the API scores candidate options in a single forward pass and returns normalized probabilities and a 0–1 confidence value to let code route uncertain cases to fallback logic or human review.
- Google publishes latency targets of roughly 150–250 ms end-to-end with median model compute of 50–100 ms, and lists input-only pricing at $0.10 per 1M input tokens with no output-token charges.
- The launch joins similar offerings from OpenAI and Microsoft and is pitched for agent routing, policy checks, content moderation and context grading, with early internal use at Google and ecosystem integrations appearing for developer adoption.