Gemini Decisions API Launches With Decision 1

Google on 11 November 2026 opened the Gemini Decisions API, a new Gemini API endpoint for closed-set decisions, powered by a new model, Gemini Decision 1 (gemini-decision-1). In a Google blog post the same day, product manager Min Choi and principal scientist Felix Yu of Google DeepMind said developers can pass one multimodal input — text, images, audio, video and PDFs, “up to 256K tokens” — “along with up to 1,000 independent questions.” The model, they wrote, “scores your candidate options in a single forward pass, returning probabilities and confidence scores in 150-250 ms end-to-end latency at $0.10 per 1M input tokens with zero output tokens billed.”

The Gemini API changelog for 11 November records the same launch: Gemini Decision (gemini-decision-1) and the Decisions API “for fast, calibrated binary, categorical, and rubric decisions across text, images, audio, video, and PDFs with zero output tokens.” Both official notices give that price on 11 November 2026. The blog adds that there is an API no-cost tier in Google AI Studio for prototyping, that the shared input is billed once per request no matter how many questions are evaluated, and that each question adds only its own instruction and criteria tokens. There are, Google says, no output-token charges, variable thinking-token bills, or separate audio-transcription or OCR costs.

The useful distinction is what this endpoint is for. It is not a general chat model. Google describes “purpose-built primitives for bounded choices like tool selection, policy checks, model routing, and context grading.” Instead of writing an answer token by token, Decision 1 returns probabilities and a confidence score from 0 to 1 for each question, so code can send ambiguous cases to a fallback or a person.

Three question types, scored in one pass

A single request can mix three question types against the same input. noul is yes/no, with optional true and false criteria. choice selects from 2 to 255 options and returns the winner plus normalised probabilities for every candidate. score grades an input on an ordered rubric starting at 0, and returns a continuous weighted score and per-level probabilities. The input field uses the same Content block format as the Interactions API. Google’s example pairs a breakroom photograph with a text note and asks all three types at once.

Google also publishes tighter latency figures as its own measurements: median model compute of 50-100 ms, with the 150-250 ms figure being end-to-end, “across multimodal inputs and contexts up to 16K tokens, remaining nearly flat from 1 to 100 questions per request.” Image and video resolution can be tuned to trade visual detail against token cost. Those numbers, like the rest of the performance case, come from Google.

The company says Decision 1 “leads across the suite, with the lowest calibration error,” on internal benchmarks covering agent routing, reasoning, ranking and multimodal understanding. It also says that, “with zero output tokens billed, Gemini Decision 1 with the Decisions API is nearly 5x cheaper across the shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API.” Both the leadership claim and the “nearly 5x cheaper” comparison are Google’s. TSN has no independent scores.

Google names four internal uses already shipping: dynamic model routing in Gemini Notebook, which it says can serve everyday queries with “up to 40% lower latency”; a rapid canvas experiment in Stitch; emoji selection in CC, from a palette of more than 200 options; and a safety layer for MCP tools in Search. Ecosystem integrations are “rolling out” across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, “with support coming soon to Gemini in Google Cloud.”

A decisions API is only as good as the options it is given

The 11 November launch is best read as Google putting a calibrated, closed-set endpoint next to its generative models, not as a replacement for them. The product’s speed and price case depends on a design choice: it never bills output tokens because it does not write an answer. That is attractive for routing, policy checks and grading, where the legal or product decision is a labelled option, not a paragraph. It is also a limit. The quality of the result still sits in the question list and the criteria the caller writes.

TSN has already covered this shape of product. OpenAI’s Decisions API and Perplexity’s Decider were earlier attempts to return a probability rather than an essay; that earlier post is cited here in the body only. IBM has also shipped decision adapters on Granite Switch 4.1, another one-pass approach to bounded choices. Google’s twist is the multimodal input, the 1,000-question batch, and a published $0.10-per-1M-input price with no output-token bill, dated 11 November. Gemini 4 Carbon’s general availability, recorded in the same changelog a day earlier, is the reasoning-model track this endpoint is meant to sit beside, not to replace: Notebook’s own use, Google says, is to reserve “deep reasoning models for complex tasks.”

The facts to watch are ordinary. Independent latency and calibration numbers would test the internal suite. So would a Google Cloud launch date, which is still “coming soon.” Until then, the Decisions API is a generally available Gemini API endpoint, with official 11 November pricing of $0.10 per 1 million input tokens and no output tokens billed, for up to 1,000 closed-set questions on one multimodal input.