Google announces Gemini Decisions API powered by Gemini Decision 1

Google says the Gemini Decisions API is powered by the model identified as `gemini-decision-1`.[2] It is designed for bounded decisions, including tool selection, policy checks, model routing and context grading.[2] The API supports yes/no questions, choices among options and rubric scoring.[2] Google says Gemini Decision 1 scores candidate options in one forward pass rather than generating answers token by token.[2] That design positions it for structured decisions where a system needs to choose or assess options, rather than produce a prose response.

Google says the API can evaluate text, images, audio, video and PDF inputs.[2] The stated multimodal input limit is up to 256K tokens.[2] A request can contain up to 1,000 independent questions, and Google says the API returns probabilities and confidence scores with its answers.[2] Those details may matter to teams considering batch evaluation or confidence-based routing.

Google reports end-to-end latency of 150–250 milliseconds and median model-compute latency of 50–100 milliseconds.[2] It says median compute latency stays nearly flat from one to 100 questions per request in contexts up to 16K tokens.[2] Google prices the API at $0.10 per million input tokens and says shared input is counted once per request, regardless of how many questions are asked.[2][3] Google also reports that the API is nearly five times cheaper across a shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API.[2] That comparison is limited to the stated benchmark suite.

Google says developers can access the API through the Gemini API in Google AI Studio.[2] Its quickstart requires SDK version 3.2 or later.[3] Google says the model generates no text tokens, leaving output-token usage at zero, and that developers can branch, sort and route directly using the API response.[3] Vercel says Gemini Decision 1 is available on AI Gateway, and that AI SDK 7 supports PDF, audio and video inputs for the model.[4] Google also says Gemini Notebook uses the API as a real-time router for model tiers and thinking levels.[5]