Skip to content

Google releases Gemini Decisions API powered by Gemini Decision 1

Google AI Studio

The endpoint evaluates classification and routing tasks across 256K-token multimodal inputs, costing $0.10 per million input tokens with no output fees.

Key points

  • Evaluates up to 1,000 questions per request across text, images, audio, video, and PDFs in a single forward pass
  • Delivers 150 to 250 ms end-to-end latency and 50 to 100 ms model compute latency without token-by-token generation
  • Provides three decision primitives: yes/no conditions, multiple-choice selection across up to 255 candidates, and ordered rubric scoring
  • Available in Google AI Studio via the Gemini API, with integrations in OpenRouter, LangChain, LiteLLM, and Pydantic AI
  • Includes sample applications such as Doorstep Parcels, Speed Sketch, and Find the Zookeeper

Why it matters

Developers can run structured classification, agent routing, and guardrails with lower latency and costs without token generation overhead or custom classifier training.

Sources