Google has launched the Gemini Decisions API, a new endpoint for fast, closed-set decisions powered by the new Gemini Decision 1 model. It evaluates up to 1,000 yes/no, choice and rubric-scoring questions in a single pass, returning probabilities and confidence scores in 150–250 ms — at a price that undercuts the general-purpose models doing the same job today.
The Gemini Decisions API is built for a specific kind of work: decisions with a closed set of answers. Instead of asking a model to write an open-ended response, you give it up to 1,000 questions — yes or no, pick one of these options, score this against a rubric — and it answers all of them in a single pass, returning probabilities and confidence scores. That is a fundamentally different shape of API call from a chat completion, designed for the machinery inside products rather than for conversation.
It works across text, images, audio, video and PDFs, with up to 256K tokens of input per call. That multimodal range matters because the decisions a product needs to make are rarely text-only: is this image appropriate, does this video match the brief, does this document meet the criteria. A decision endpoint that can judge all of those in one pass is a tool for building products that need judgement at scale.
The headline number is the latency: 150–250 ms end-to-end. For a decision gating a user-facing interaction — which model should answer, is this content safe, which result ranks first — that speed is the difference between a product that feels instant and one that feels sluggish. Google is pricing this endpoint to be used constantly, not sparingly.
The API is powered by Gemini Decision 1 (gemini-decision-1), a new model built specifically for this job, available through the Gemini API in Google AI Studio. Google says Gemini Decision 1 leads across benchmarks spanning agent routing, reasoning, ranking and multimodal understanding, with the lowest calibration error — its confidence scores are the most honest about how likely it is to be right. That matters because products act on the confidence, not just the answer.
Pricing is $0.10 per 1M input tokens with zero output tokens billed, plus a no-cost tier in Google AI Studio, and Google says it is nearly 5x cheaper across the shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API. The zero-output-tokens detail is the clever part: the API returns probabilities and scores rather than generated text, so there is no output to bill for — the whole cost of a decision is the input, and the input is cheap.
That pricing is a direct challenge to the current default: routing every small judgement through a general-purpose model at general-purpose prices. The Decisions API is Google's argument that the judgement layer of AI products deserves its own specialised, cheap, fast endpoint — and the companies who adopt it will make far more decisions, far more cheaply, than the ones who do not.
Google is not launching this as a theory. Internal uses are already live: dynamic model routing in Gemini Notebook, which cut latency by up to 40% on everyday queries; rapid design iteration in Stitch; emoji selection in CC, a Google Labs experiment; and a safety layer for MCP tools in Search. The Notebook example is the most telling — the routing decision itself stopped being the bottleneck, so the whole product got faster.
The ecosystem integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode. That list is the developer stack for building AI products, and Google is putting the Decisions API into all of it at once: this is a building block that will be available wherever you already build.
The pattern here is one the AI industry keeps repeating: the big, general models get the headlines, and the small, specialised models do the work that makes products viable. A decisions endpoint is infrastructure, not a demo — and infrastructure is what determines which products can afford to exist.
For SEOs, the Decisions API is a window into how Google's own products make judgements — and, by extension, how the systems that evaluate your content might work. Agent routing, ranking and rubric scoring are all decisions of exactly the kind this API handles, and the fact that Google is building a specialised model for them tells you how seriously it takes getting those decisions right, fast and cheaply. Content that is easy to judge — clear, structured, unambiguous — will score better in any system that judges it this way.
For link builders and tool builders, the API is a practical opportunity. SEO tools make thousands of small judgements: is this link relevant, does this page match this query, is this prospect a good fit. Doing those judgements with a cheap, fast decision endpoint instead of an expensive general model changes the economics of building a tool.
The broader lesson is about where AI search is heading. The future is not one giant model doing everything; it is a pipeline of specialised models, each doing one job well, with decision models like Gemini Decision 1 routing between them. Understanding that pipeline — and building content that survives each stage of it — is the new core skill for anyone who wants to be visible in AI search.
Sources: Google The Keyword blog, November 11, 2026.
Related reading on this site: Gemini 4 Hydrogen Is Now the Default Model in AI Mode for Everyone, Google: Technical SEO Remains Key for AI SEO, Direct Answers Win AI Citations: 193,507 Answers Studied, YouTube Takes 24.5% of AI Overview Citations in Ahrefs' November Ranking, Google Explains How Redirects Affect Ranking Signals, and November Search Console Click Spikes Are Bot Traffic.
Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.
Start free trial