Google introduced the Gemini Decisions API on November 11, powered by gemini-decision-1. It is available through the Gemini API in Google AI Studio.

Bounded choices

The endpoint returns probabilities, confidence or rubric scores over multimodal input instead of generating chat text. Google describes up to 256K input tokens and 1,000 independent questions per request.

A distinct billing model

Google lists $0.10 per million input tokens with no output-token charge. Shared input is billed once; each question adds instructions and criteria.

The advertised 150–250 ms end-to-end latency is a vendor result, not a guarantee for every request. Confidence values also need validation for consequential decisions. No general language-model leaderboard mapping is inferred.