Google built an AI model that can’t write a sentence — and that’s the point

The article covers Google's new Gemini Decisions API, which uses a model called Gemini Decision 1 to choose among fixed options rather than generating a written response. The model is pitched as faster and cheaper than general generative models for the routine decisions AI apps make constantly, with reported response times of 150 to 250 milliseconds and pricing starting at $0.10 per million input tokens. The piece places the launch in a growing market that already includes offerings from OpenAI, TypeSafe AI, and Microsoft. Google says it already uses the model internally, for routing in Gemini Notebook and for spotting adversarial behavior around MCP tools in Search, but it has not published independent evaluations. Readers will also learn the benchmark caveats: Google's model edged out rivals on its own test, yet a competitor was cheaper to run and won on two of seven shared tests. Billing works by charging for shared context once, though each question still counts toward tokens, and the API is available now through AI Studio, with Google Cloud support planned but undated.