Gemini Decisions API
First reported by Blog.google ·
Automated content moderation and data analysis tasks become significantly faster and cheaper.
Google has launched the Gemini Decisions API, powered by its new Gemini Decision 1 model. This API allows for fast, probabilistic decision-making across multiple data types including text, images, audio, video, and PDFs. It can evaluate up to 1,000 yes/no, multiple-choice, or rubric-scoring questions in a single pass. The model scores candidate options directly, offering high speed with an end-to-end latency of 150-250 milliseconds. This approach is designed to be cost-effective, with pricing based on input tokens and zero output tokens billed, making it approximately five times cheaper than using the Interactions API for similar tasks. Google DeepMind product managers Min Choi and Felix Yu announced the API, highlighting its potential for applications like dynamic model routing, content grading, and policy checks.
The Gemini Decisions API represents a shift towards specialized AI models optimized for specific, high-throughput tasks rather than general-purpose generation. By processing multimodal inputs and executing bounded choices in a single pass, it significantly reduces latency and cost for applications requiring rapid evaluation of conditions or categorization. This is particularly beneficial for real-time systems where milliseconds matter, such as content moderation, routing user queries, or grading user-generated content at scale.
This development impacts developers building applications that rely on immediate, consistent decision-making from AI. The API's ability to handle multiple question types and modalities simultaneously offers a streamlined alternative to custom classifiers or sequential generative AI calls. Companies looking to implement features like instant emoji reactions, dynamic model routing in chatbots, or efficient safety checks within search tools can now achieve these with greater speed and cost efficiency.
AI-written summary. May contain errors.