Google introduced the Gemini Decisions API, a specialized service designed to automate structured business decision-making at $0.10 per one million input tokens.
Unlike open-ended generative AI, the Decisions API returns discrete answers paired with confidence scores—quantifying how certain the Gemini Decision 1 model is in each output. This design targets fraud detection, loan underwriting and customer service routing, where enterprises need both accuracy and auditability to satisfy compliance teams.
The pricing undercuts general-purpose LLM alternatives. For a bank processing 10 million underwriting decisions monthly, the $1,000 monthly cost represents a meaningful arbitrage against building custom models or buying point solutions from specialized vendors like Sift or Emailage. The consumption-based structure also appeals to enterprises skeptical of large platform commitments.
This move reflects a wider market shift: AI vendors are fragmenting from "one model to rule them all" into task-specific models with clearer ROI stories. OpenAI has gone this route with its Reasoning models. Anthropic is packaging Claude for specific use cases. Google is following the same playbook—lock enterprises into Google Cloud infrastructure by solving their most measurable problems first.
The Decisions API also lets Google collect decision-making data at scale. That signals a longer bet: train future versions of Gemini on the patterns of how enterprises actually make decisions, building a competitive moat that pure-play API vendors cannot match.
