On November 11, 2026, Google DeepMind announced the Gemini Decisions API, a new endpoint for fast, closed-set decisions such as tool selection, policy checks, model routing and grading. It runs on a new model, Gemini Decision 1 (gemini-decision-1). Instead of writing an answer, Google says it scores the candidate options in a single forward pass and returns probabilities and a confidence score.
What to know
- Three question types:
noul(yes/no, returns a probability),choice(2 to 255 options, returns the winner and a probability for each) andscore(an ordered rubric, returns a weighted score and per-level probabilities). Every answer includes a confidence score from 0 to 1. - Many questions, one input: one request can carry up to 1,000 independent questions about one shared input. Google’s docs list a 262,144-token context window across the input and all questions. Inputs can be text, images, audio, video and PDFs.
- Price: Google lists $0.10 per million input tokens and $0 for output tokens, with no thinking-token charges. The shared input is billed once per request, however many questions you ask. There is a no-cost tier in AI Studio for prototyping.
- Speed (Google’s figures): 150 to 250 ms end to end, with 50 to 100 ms median model compute. Google says latency stays nearly flat from 1 to 100 questions for contexts up to 16K tokens.
- Where: in the Gemini API through Google AI Studio, on the
v1betaendpoint. Google says support in Google Cloud is “coming soon”, and integrations are rolling out for OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode. - Not supported yet (per the docs): context caching, the Batch API, Flex or Priority inference, YouTube URLs and
gs://URIs. - Claims to weigh: Google says Gemini Decision 1 “leads across the suite, with the lowest calibration error” on its own benchmark suite, and is nearly 5x cheaper there than Gemini 3.5 Flash-Lite with the Interactions API. These are Google’s figures.
- Status: neither the blog post nor the docs label the API as preview or generally available.
Why it matters
Routing, moderation and grading steps in an agent are usually small yes/no or pick-one questions, and a full chat model is slow and costly for them. A scoring endpoint with a confidence number lets your code send unclear cases to a person or a bigger model. OpenAI has a similar product: see our note on the OpenAI Decisions API. For how to word the questions, read our Gemini Decisions API question guide, and for when to stop and ask a human, the agent permissions checklist. Background on the Gemini page and in the model tracker.
Source: Google. We summarise in our own words; check the source for the full details.