Google has launched the Gemini Decisions API, a new endpoint powered by its Gemini Decision1 model that helps AI agents make fast, preset decisions without generating lengthy text responses.
The interface allows developers to evaluate up to 1,000 independent questions against shared multimodal inputs that include text, images, audio, video, and PDFs. Rather than generating replies token by token, Gemini Decision1 scores preset options in a single forward pass and returns probabilities along with confidence scores, with end-to-end latency of 150–250 milliseconds.
Google states that the model is purpose-built for bounded tasks, such as selecting tools, checking policy compliance, routing queries to the correct model, and evaluating context. Developers can submit yes-or-no checks, multiple-choice questions with up to 255 options, or scoring requests based on preset rubrics in a single API call.
Faster and cheaper decision-making
According to Google, the API has a median model computation time of 50–100 milliseconds, and performance remains nearly unchanged when processing 1 to 100 questions per request. The model supports a context window of up to 256,000 tokens and can handle multiple types of data simultaneously. This allows developers to combine inputs such as audio recordings, images, and documents without relying on separate processing systems.
Google also claims that the model outperforms rival systems on benchmarks covering agent routing, reasoning, ranking, and multimodal understanding, while achieving the lowest calibration error. The company says that in its shared benchmark suite, Gemini Decision1 is nearly 5 times cheaper than Gemini 3.5 Flash-Lite using the Interactions API.
The release comes amid increasingly intense competition in specialized AI decision-making tools. OpenAI launched its Decisions API in October, powered by its GPT-6 Luna model. The company says the service makes decisions 10 times faster than its Responses API and supports up to 400 questions per request. Like Google's offering, it provides probability-based checks, multiple-choice selection, and rubric-based scoring, with pricing starting at $0.10 per million input tokens.
The Decisions API is priced at $0.10 per million input tokens, with no charge for output tokens. Regardless of how many questions are evaluated, developers pay only once for the shared input data, although each question adds tokens related to instructions and criteria.
Powering real-world AI applications
Google has already deployed this technology across multiple products, demonstrating how specialized decision models can improve the speed and efficiency of automated systems. In Gemini Notebook, the API routes incoming queries to the appropriate model tier and reasoning level. According to Google, this approach can reduce latency for routine queries by as much as 40% while reserving more powerful reasoning models for complex tasks.
The technology also powers rapid design experimentation on Google's Stitch platform, allowing users to generate interactive product wireframes and make changes through voice commands or touch input. Meanwhile, Google's CC assistant uses the API to pick emoji reactions from more than 200 options in a fraction of a second. Google Search is also using this technology as a safety layer to detect adversarial behavior involving Model Context Protocol (MCP) tools. Beyond these applications, the API can help developers build systems that automatically select appropriate tools, evaluate incoming information, and determine when human intervention is needed.
The Decisions API is now available through Google AI Studio, which includes a free tier for prototyping. Google is also rolling out integrations with developer platforms such as OpenRouter, LangChain, Vercel, LiteLLM, and Pydantic AI.
Comments