Google prices a model call for routing and policy checks

Google released an endpoint that answers closed questions and returns no generated text. A request carries one input of up to 256K tokens and up to 1,000 independent questions. Each question is a yes/no check, a choice among 2 to 255 options, or a score on an ordered rubric, and each answer comes back with a confidence value from 0 to 1. Google quotes 150 to 250 ms end to end for contexts up to 16K tokens, and a price of $0.10 per million input tokens with no charge for output. The post lists tool selection, policy checks, model routing, and context grading as intended uses, and says Google Search uses the endpoint as a safety layer for MCP tools. Integrations are "rolling out" to OpenRouter, LangChain, LiteLLM, OpenCode and others.

Google says the model, Gemini Decision 1, leads its benchmark suite with the lowest calibration error and is "nearly 5x cheaper" than Gemini 3.5 Flash-Lite on the same suite. Those are Google's own measurements.

Why it matters: A harness makes many small decisions per task: which tool, which model tier, whether an action breaks a rule. A generative call for each one adds latency and cost. A cheap endpoint with a confidence value lets the code send low-confidence cases to a slower model or a person. The calibration claim is the one to test on your own traffic before a gate depends on it.