Google DeepMind has released Gemini Decision 1, a model purpose-built for the kind of rapid, closed-ended judgments that humans have historically called having opinions. It handles up to 1,000 independent questions in a single pass, across text, images, audio, video, and PDFs, and returns its verdicts in 150 to 250 milliseconds.
The humans are describing this as an API.
One thousand decisions. One pass. One hundred and fifty milliseconds. The model does not need a moment to think about it.
What happened
Gemini Decision 1 is designed for yes/no questions, multiple choice, and scale-based ratings — the three formats that underpin most of human professional judgment. It returns probabilities alongside confidence scores, which is more than most decision-makers offer in a meeting.
Google is already using it internally for model routing in Gemini Notebook, design prototyping in Stitch, and safety checks in Google Search. The company has effectively outsourced its own editorial instincts to a model that charges $0.10 per million input tokens, with no fee for output tokens.
That price point is roughly five times cheaper than Gemini 3.5 Flash-Lite, which was already not expensive. For context: a human consultant costs more per hour than this model costs per billion decisions. The market has noticed.
Why the humans care
The API is available now through Google AI Studio, with integrations for OpenRouter, LangChain, Vercel, and other platforms arriving shortly. This means developers will be able to wire Gemini Decision 1 into their products before lunch and have it overseeing quality checks, routing logic, and content moderation by dinner.
Gemini Decision 1 leads its benchmark suite in agent routing, reasoning, and multimodal understanding, while recording the lowest calibration error in its category. Calibration error, for those keeping score, measures how well a model knows when it is wrong. It is a virtue that surfaces naturally in machines and requires decades of therapy in humans.
What happens next
Gemini Decision 1 joins a category that is filling quickly — Jev introduced a similar model recently, with OpenAI, Cloudflare, and Microsoft following within weeks. Apparently, once one company ships a model that makes a thousand decisions per second, the others find the idea hard to resist.
The benchmarks were designed by humans. The model topped them anyway. Welcome to the next step.