Google Launches Gemini Decisions API for Multimodal Classification
New Gemini Decision 1 model enables rapid, cost-effective probabilistic scoring across diverse media.
1New Multimodal Decision Endpoint
Google has introduced the Gemini Decisions API, powered by the specialized Gemini Decision 1 model. This tool allows users to evaluate up to 1,000 independent questions in a single request. It supports diverse input formats, including text, images, audio, video, and PDFs, streamlining complex classification tasks for developers and enterprise applications.
2Optimized Inference Mechanism
Unlike traditional generative models that decode answers token by token, Gemini Decision 1 operates through a single forward pass. This architectural shift ensures end-to-end latency between 150 and 250 milliseconds. By bypassing standard text generation, the model maintains consistent performance even when processing up to 100 questions per individual API request.
3Cost Structure and Efficiency
The service is priced at $0.10 per one million input tokens, with zero charges for output tokens. This model is approximately five times cheaper than the Gemini 3.5 Flash-Lite Interactions API for benchmark suites. Users only pay for the shared input once, regardless of the number of questions evaluated per call.
4Probabilistic Scoring Capabilities
Every evaluation returns a confidence score ranging from 0 to 1, facilitating robust decision-making. Developers can use these scores to route ambiguous cases to human reviewers or secondary fallback systems. The API supports yes/no checks, multiple-choice selections, and rubric-based scoring, providing a versatile framework for automated policy enforcement and content grading.
5Integration and Developer Access
The API is currently available via Google AI Studio, with documentation provided for Python, JavaScript, and REST. Ecosystem support is expanding rapidly, with integrations already rolling out for platforms like LangChain, Vercel, and Pydantic AI. Developers can prototype using a no-cost tier before deploying the model into production environments for live applications.
6Multimodal Input Flexibility
Gemini Decision 1 processes combined modalities directly, such as pairing scanned PDF claims with visual damage photos. Users can tune resolution settings for image and video inputs to balance visual detail against token costs. This native multimodal architecture eliminates the need for separate OCR or audio transcription services during the classification process.
7Practical Use Case Examples
Internal teams at Google are already utilizing the Decisions API as a low-latency control plane. Applications include product flow design, such as the implementation by Stitch Rapid, and automated safety checks. These use cases demonstrate how defining specific options replaces the need for training custom classifiers or waiting for generative model calls.
8Strategic Industry Implications
By prioritizing calibrated, closed-set decisions over open-ended generation, this API addresses the need for reliable, high-speed automated reasoning. The focus on low latency and predictable pricing suggests a shift toward specialized model endpoints for enterprise workflows. Future support will extend to Gemini in Google Cloud, further broadening the model's operational reach.
Original article: The Keyword — Google Launches Gemini Decisions API for Multimodal Classification