Gemini Decisions API Scores Bounded Choices in Milliseconds
Google’s Gemini Decisions API is a specialized endpoint for fast classification of bounded choices. Unlike text-generating models that process sequentially, Gemini Decision 1 evaluates options in a single computational pass, delivering probabilities and confidence scores in 150-250 milliseconds. It processes multimodal inputs—text, images, audio, video, and PDFs totaling up to 256K tokens—and can evaluate up to 1,000 independent questions per request. Pricing is $0.10 per million input tokens with zero output token charges.
The model supports three decision types: binary yes/no assessment, multiple-choice selection from 2 to 255 options, and rubric-based scoring across ordered categories. Each result includes a confidence score that allows systems to escalate ambiguous decisions to fallback mechanisms or human review. In benchmark tests covering agent routing, reasoning, ranking, and multimodal comprehension, Gemini Decision 1 achieves the lowest calibration error and costs nearly five times less than Gemini 3.5 Flash-Lite for equivalent classifications.
Google’s internal teams are already deploying it as a control layer. Gemini Notebook routes queries to optimal model tiers using the API, achieving up to 40 percent latency reduction on everyday queries. Stitch generates product wireframes in seconds. CC from Google Labs selects emoji reactions from a palette exceeding 200 options in fractions of a second. Google Search applies it to detect adversarial content. The API is available through Google AI Studio with a free tier for development, and integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit, and OpenCode.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.