By Hamza Faisal
Google has released a model that never writes a word. Gemini Decision 1, announced on November 11 by Google DeepMind, powers the new Gemini Decisions API — an endpoint that answers up to 1,000 yes-or-no, multiple-choice or rubric questions about a piece of content in a single pass, returning probabilities and confidence scores instead of prose.
The idea is aimed at the plumbing of AI agents rather than the chat window. Instead of asking a model to write an answer and then parsing the text back into a structured value, developers send one multimodal input — text, images, audio, video or a PDF, up to 256,000 tokens — along with a list of questions, and get back a set of scored answers.
Built for the decisions agents make all day
Google pitches the API at the bounded choices software makes constantly: which tool an agent should call next, whether a piece of content passes a policy check, which model tier should handle a query, how to grade a response, or how to rank a list of options.
Speed is the selling point. Google says the model scores its options in a single forward pass, with median model compute of 50 to 100 milliseconds and 150 to 250 milliseconds end to end, and latency stays nearly flat whether a request carries one question or 100.
Pricing follows the same logic: $0.10 per million input tokens, with nothing charged for output — because there are no output tokens. The shared input is billed once per request no matter how many questions are asked. A no-cost tier in Google AI Studio lets developers prototype.
Google is already using it inside its own products
The company says Gemini Notebook already uses the API as a real-time router for model tiers and thinking levels, cutting latency by up to 40 percent on everyday queries while reserving deeper reasoning models for harder ones. It is also being used in a Stitch design experiment and as a safety layer for MCP tools in Search.
The launch lands a day after Google made Gemini 4 Carbon generally available, capping a busy week for DeepMind that has also included AlphaProof Code and Gemini Robotics 2.5.
A crowded race to answer, not write
Google is not first to the idea. TypeSafe launched Jev in September, OpenAI unveiled its own Decisions API at DevDay on September 29, and Microsoft followed with Microsoft-Decision-1, built on Alibaba's Qwen. Google's own benchmarks put Decision 1 ahead of Jev 1.14 on five of seven shared tests, and Artificial Analysis scores it 66 on its Decision Index — but those figures are Google's own and have not been independently verified.
There are limits at launch. Access runs through the Gemini Developer API v1beta, Google Cloud support is still coming, and features such as prompt caching and the Batch API are not supported yet. Integrations with OpenRouter, LangChain, Vercel and LiteLLM are rolling out.
Even so, the direction is clear: as agents take on more of the work, the model that picks the right option quickly and cheaply may matter as much as the one that writes the best paragraph. Gemini Decision 1 is Google's bet that the next frontier for AI is not generation — it is judgement.



