Google launches Gemini Decisions API for fast, low-cost multimodal choices

According to Google AI:

How it works

Instead of generating answer text token by token like traditional language models, Gemini Decision 1 scores candidate options in a single forward pass. It supports text, images, audio, video, and PDFs up to 256K tokens, allowing developers to ask up to 1,000 independent questions simultaneously. Every answer includes a confidence score, letting developers route ambiguous cases to fallback systems or human reviewers.

Pricing and Performance

The model delivers a median latency of 150-250 ms end-to-end and is priced at $0.10 per 1M input tokens with zero output tokens billed. Google reports that it is nearly 5x cheaper than standard generative alternatives across shared benchmark suites.

Real-World Applications

Internal teams at Google are already using the API as a low-latency control plane. Examples include dynamic model routing in Gemini Notebook to reduce query latency by up to 40%, rapid design iteration in Stitch, instant emoji selection in Google Labs, and safety layers for Model Context Protocol tools in Google Search.