Google’s new Gemini Decision 1 model does not generate text. Instead, it makes yes-or-no, multiple-choice and scoring decisions in 150 to 250 milliseconds.
Google has introduced the Gemini Decisions API, a new interface designed for fast decisions with predefined response options. It is powered by the new Gemini Decision 1 model. Developers can submit an input containing text, images, audio, video or PDF files, with a context window of up to approximately 256,000 tokens, and ask up to 1,000 independent questions about it.
“Rather than generating answer text token by token, Gemini Decision 1 scores your candidate options in a single forward pass.”
Three Types of Decisions
The API supports yes-or-no questions, multiple-choice questions with between 2 and 255 options, and ratings based on a predefined scale. Instead of generating text, it returns probabilities and a confidence score between 0 and 1. Applications can use these values to route uncertain cases to human reviewers or another process. According to Google, response times range from 150 to 250 milliseconds. The model is not designed for open-ended text generation, creative writing or complex reasoning.
Free Output Tokens
Usage costs $0.10 per million input tokens, while output tokens are free. The shared input is billed only once, regardless of how many questions are asked about it. In Google’s own benchmarks, this makes the solution almost five times cheaper than its Gemini 3.5 Flash-Lite model. However, independent tests are still needed to determine whether the confidence scores remain reliable when working with real-world enterprise data.
Google identifies several potential use cases, including AI agents selecting tools, policy checks, routing requests to suitable models and content evaluation. Google is already using the API internally. Among other applications, Gemini Notebook uses it to route requests to the appropriate model. According to Google, this has reduced response times for everyday queries by up to 40 percent. Google Search also uses the API as a protective layer for MCP tools.
The API is available in beta through the Gemini API in Google AI Studio, including a free usage tier. Google Cloud support is expected to follow. Platforms and frameworks such as OpenRouter, LangChain, Vercel, LiteLLM and Pydantic AI already support the interface.
(Editorial Team)