Google has released the Gemini Decisions API, an endpoint for closed-set decisions that returns probabilities instead of generated text. Google announced it on its developer blog on November 11. It runs on a new model, Gemini Decision 1 (gemini-decision-1), and is available now through the Gemini API and Google AI Studio.
Instead of writing an answer token by token, the model scores the candidate options in a single forward pass. It supports three question types: yes/no, which estimates the probability that a condition holds; multiple choice, with 2 to 255 options and normalized probabilities; and scoring against an ordered rubric, which returns a weighted score. One request can carry up to 1,000 questions over text, images, audio, video and PDFs, with up to 256K tokens of input. Every answer comes with a confidence score from 0 to 1.
Google reports median model compute of 50 to 100 milliseconds, or 150 to 250 milliseconds end to end, and says latency stays nearly flat from 1 to 100 questions per request. Pricing is $0.10 per million input tokens with no charge for output. Shared input is billed once per request however many questions use it, and thinking tokens, audio transcription and OCR are not billed separately. Google says that on its benchmark suite the API is nearly five times cheaper than Gemini 3.5 Flash-Lite with the Interactions API. A free tier is available in AI Studio.
Google says it already uses the model in its own products. Gemini Notebook uses it to send each query to the right model tier, with up to 40% lower latency for everyday queries, and Google Search uses it as a safety layer to detect adversarial behavior in MCP tools. Support is rolling out in OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, and Google says Google Cloud support is "coming soon."
For teams building agents, this targets the many small classification calls in a pipeline: routing, guardrails, moderation and ranking. These are often handled by a full generative model call and a parser. Because the API returns calibrated probabilities, code can send low-confidence cases to a stronger model or a human reviewer without extra prompting.