Google launched the Gemini Decisions API on November 11, 2026, providing software with bounded choices and probabilities without generating text. Powered by Gemini Decision 1, it supports binary questions, supplied choice lists and ratings against predefined scales. A request can combine multiple content types and include up to 1,000 questions, within a total limit of 262,144 tokens, with reported latency of 150 to 250 milliseconds for contexts of up to 16,000 tokens. Google charges $0.10 per million input tokens, with no output-token charge. The company recommends testing confidence thresholds and retaining human or second checks for high-stakes decisions, such as payouts.
Source: Read the original article


