Gemini Decisions API: Fast, calibrated multimodal decisions

Practical AI: Tools, Models & Frameworksmultimodalapi

What changed

Google introduced the Gemini Decisions API with a new Gemini Decision 1 model, a specialized endpoint for bounded classification tasks (yes/no, multiple choice, rubric scoring) that scores candidate options in a single forward pass instead of generating text token-by-token. It processes multimodal inputs (text, images, audio, video, PDFs up to 256K tokens) with up to 1,000 independent questions per request, returning results with confidence scores in 150-250 ms end-to-end latency, priced at $0.10 per 1M input tokens with zero output token charges.

Why it matters

Builders can replace token-streaming classifiers with a purpose-built decision primitive that is 50-100x faster at inference (median 50-100 ms compute) and ~5x cheaper than general-purpose models for routing, policy checks, and classification tasks. The no-output-token billing and flat latency across question counts make it practical for real-time control planes in production systems.

How it compares

Unlike Gemini 3.5 Flash or other generative models that must decode answers sequentially, Decisions API scores fixed candidate sets in one pass, making it fundamentally faster and cheaper for closed-set problems. It differs from the prior open-source AnyDecisionModel announcement by being a commercial API-only service from Google focused on cloud deployment, with native multimodal support built on Gemini's architecture rather than edge-deployable models.

Sources