Google DeepMind is adding a decision-focused option to its AI offerings. The Gemini Decisions API is designed for tasks with a defined set of answers, such as yes-or-no questions, multiple choice, and ratings on a scale.
At its core is Gemini Decision 1, which can evaluate up to 1,000 independent questions in one pass. The approach is aimed at situations where a system needs to score choices quickly, rather than generate a long open-ended response.
One pass for many types of input
Gemini Decision 1 accepts text, images, audio, video, and PDFs. For each question, it scores the available options and returns probabilities alongside confidence values. Google says responses arrive in 150 to 250 milliseconds.
Handling many separate questions together may suit workflows that repeatedly make the same kind of judgment. A team could, for example, need to sort or assess a batch of material against fixed criteria. The API is built around those bounded choices, rather than requiring a detailed written answer for every item.
The multimodal support means the inputs do not have to be text alone. Images, audio, video, and documents are also within the stated range, giving developers a way to apply closed-ended scoring across different formats.
Google says it is already using the API
Google reports using the API internally for several purposes. These include dynamic model routing in Gemini Notebook, design prototyping in Stitch, and safety checks in Google Search.
Those examples point to different roles for a decision model. Model routing involves selecting between options, while design prototyping and safety checks can involve assessing material against criteria. The common thread is a task that can be framed as a choice or rating.
The API is available through Google AI Studio. Integrations for OpenRouter, LangChain, Vercel, and other platforms are rolling out, which could make the service accessible through a wider set of developer tools.
Pricing puts the focus on input volume
Google lists a price of $0.10 per million input tokens and says there is no output-token charge. The company says this is about five times cheaper than Gemini 3.5 Flash-Lite.
That pricing structure is relevant to a model designed to process many questions in a single pass: the stated fee is tied to input tokens, while generated output does not add a separate token charge. Actual cost depends on the amount of input supplied, so the per-million rate is the key figure developers can use when considering batches of questions.
A growing category of decision models
Gemini Decision 1 arrives amid a recent wave of similar offerings. The source article says startup Jev introduced a version first, followed quickly by OpenAI, Cloudflare, and Microsoft with comparable products.
The interest in this category reflects a specific need: not every AI task calls for a conversation or a free-form explanation. Some workflows need a system to compare choices, assign a rating, or make a quick yes-or-no assessment. Google’s API packages those jobs as direct scoring, with probabilities and confidence values returned for review.
For developers, the distinction is practical. A decision API is intended for questions with a clear answer space, and its value depends on whether a task can be expressed in those terms. Gemini Decision 1 combines that format with batch processing, multiple input types, and a stated response time of 150 to 250 milliseconds.