Gemini Decisions API scores up to 1,000 bounded questions in one request

Google has launched the Gemini Decisions API for applications that need a bounded choice rather than generated text. One request can apply up to 1,000 yes-or-no, multiple-choice or ordered-rubric questions to shared text, images, audio, video or PDF input.

The endpoint runs Gemini Decision 1 in a single forward pass and returns option probabilities and confidence scores. Google reports 150–250 milliseconds of end-to-end latency and nearly flat model-compute latency from one to 100 questions. Inputs can reach 256,000 tokens, although Google’s published latency benchmark covers contexts up to 16,000 tokens.

Pricing is $0.10 per million input tokens with no output-token charge. Google says the model was nearly five times cheaper than Gemini 3.5 Flash-Lite on its shared benchmark suite. Those are provider measurements, so developers still need to test calibration and failure modes on their own routing, grading or policy tasks.

Google already uses the API to route requests in Gemini Notebook, classify interactions in Stitch and check adversarial behavior around Search tools. Developers can access it through Google AI Studio and Python, JavaScript or REST interfaces. Integrations with OpenRouter, LangChain and several other frameworks are rolling out, while Gemini support in Google Cloud is listed as coming later.