Google on November 11, 2026 introduced the Gemini Decisions API, a new endpoint in the Gemini API built for fast, closed-set judgments rather than open-ended text. It is powered by a new model, Gemini Decision 1 (model ID gemini-decision-1), and is available now through Google AI Studio, including a no-cost tier for prototyping. Google pitches it for the small, frequent choices that sit inside AI products and agents: picking a tool, checking a policy, routing a query to the right model, or grading a piece of context.
The core idea is that the model never writes an answer token by token. A developer sends one shared input, which can mix text, images, audio, video and PDFs up to roughly 256,000 tokens, along with as many as 1,000 independent questions. Gemini Decision 1 scores the candidate answers in a single forward pass and returns probabilities plus a confidence score from 0 to 1 for each question. Three question types are supported: a yes/no check, a multiple-choice pick from 2 to 255 options, and an ordered rubric that returns a weighted score. Google says the confidence values let applications send ambiguous cases to a fallback model or a human reviewer.
Pricing is $0.10 per million input tokens and nothing for output, since no output tokens are generated. The shared input is billed once per request regardless of how many questions are asked; each question adds only its own instruction and criteria tokens. Google reports end-to-end latency of 150 to 250 milliseconds for inputs up to 16,000 tokens, staying nearly flat from one to 100 questions per request, and says the setup is nearly five times cheaper on its benchmark suite than running Gemini 3.5 Flash-Lite through the standard Interactions API. These figures are self-reported, and VentureBeat cautions that teams planning to use the full context window or the 1,000-question ceiling should measure latency on their own files.
Google says the API is already in use internally: Gemini Notebook uses it to route incoming queries to the right model tier and thinking level, with up to 40 percent lower latency on everyday queries; Google Search uses it as a safety layer for MCP tools; and Google Labs projects use it for instant emoji selection and rapid wireframe editing in Stitch. Integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, with support in Gemini in Google Cloud listed as coming soon.
Google is entering a category that already has competitors. VentureBeat, SiliconANGLE and developer Simon Willison all note that OpenAI's Decisions API, built on GPT-6 Luna, and TypeSafe's Jev already offer similar yes/no, choice and rubric scoring; Willison points out that Google even reuses Jev's term "noul" for a yes/no question. OpenAI matches Google's $0.10 base input rate, while Jev is cheaper at about 4.2 cents per million input tokens but handles text and JSON with only opt-in image support. Gemini's clearest differentiator is native audio, video and PDF input in the same request.
On benchmarks, the picture depends on who is measuring. In Google's launch comparisons, as summarized by VentureBeat, Gemini Decision 1 led five of the seven tests all four compared models ran and posted the lowest calibration error, about half of Jev's. On the independent Artificial Analysis Decision Index v2.1, VentureBeat reports Gemini Decision 1 placed ninth of 18 models at launch, close to Jev on overall score but behind on time per task and cost per task. SiliconANGLE separately describes Google as claiming wins on seven of nine benchmarks; DigiEditorial has relied on Google's published materials and flags that third-party tallies differ.