Signal

Google launches Gemini Decisions API, powered by its new Gemini Decision 1 model, to return predefined answers with confidence scores, at $0.10/1M input tokens

First reported by Blog.google ·

The signal ●●●○ Compiled by AI from Blog.google and Techmeme
Why you might care

The cost of making AI-powered classification and decision-making drops by an order of magnitude for many tasks.

What happened

Google has launched the Gemini Decisions API, powered by its new Gemini Decision 1 model. This API is designed for fast, closed-set decisions, allowing users to evaluate up to 1,000 yes/no, multiple-choice, or rubric-scoring questions in a single pass across multimodal inputs like text, images, audio, video, and PDFs. The model scores candidate options directly without token-by-token decoding, achieving end-to-end latency of 150-250 milliseconds. The API is priced at $0.10 per 1 million input tokens and bills zero output tokens, making it significantly cheaper than standard generative APIs for specific decision-making tasks. It provides confidence scores for each answer, enabling fallback routing for ambiguous cases. Google teams are already leveraging this API for applications such as dynamic model routing in Gemini Notebook, rapid wireframe generation in Stitch, and instant emoji selection in CC.

What it means

The Gemini Decisions API introduces a specialized, cost-effective approach for AI-driven decision-making, moving beyond traditional generative models for tasks like content grading, policy checks, and model routing. By processing multiple questions across various data types in a single, rapid pass with high confidence scores, it enables developers to build more efficient and responsive applications. This specialized model is optimized for speed and accuracy in closed-set scenarios, differentiating itself from more general-purpose LLMs.

This launch signals a growing trend toward specialized AI models tailored for specific, high-volume tasks, rather than relying solely on general-purpose LLMs for everything. The significantly lower cost and faster latency for decision-making tasks could accelerate the adoption of AI in workflows requiring quick, programmatic evaluations. Developers will need to consider how to best integrate these specialized APIs into their existing systems to leverage these performance and cost benefits.

AI-written summary. May contain errors.

Google Crypto