Google's New AI Doesn't Write Answers — It Decides: Gemini Decision 1 Scores 1,000 Questions in One Pass

A round road sign with a large white arrow pointing straight ahead, painted in bright diagonal stripes, above a city street with tram tracks

Photo: Hippopx (public domain). Illustrative image.

What happened. Google launched the Gemini Decisions API on Tuesday 11 November 2026, powered by a new model, Gemini Decision 1. The pitch is narrow by design: most AI models write answers word by word. This one does not write at all. It reads an input, looks at a fixed list of options, and tells you which option fits — and how confident it is — in a single pass.

How a model answers without writing. Developers send one shared input of up to 256,000 tokens — text, images, audio, video or PDFs — alongside up to 1,000 independent questions about it. Google’s documentation describes three question types: yes/no, a choice among 2 to 255 options, or a grade on an ordered rubric scale. Gemini Decision 1 scores every candidate option at once, in a single forward pass, and returns probabilities plus a calibrated confidence score from 0 to 1. Google puts the model compute at 50 to 100 milliseconds, 150 to 250 milliseconds end to end, and says that figure stays nearly flat from 1 to 100 questions per request at contexts up to 16,000 tokens. The intended uses are the small, repeated judgements that AI agents make all day: picking a tool, checking a policy, routing a query to the right model, grading context.

The price is the point. Gemini Decision 1 costs $0.10 per million input tokens, with no output tokens to pay for — there are none, because the model writes nothing. A no-cost tier in Google AI Studio is available for prototyping. Google also says the Decisions API is nearly five times cheaper than using Gemini 3.5 Flash-Lite through its Interactions API on its shared benchmark suite — a Google figure, not an independent one.

Google’s scorecard — and who it leaves out. Google’s announcement compares Gemini Decision 1 directly with TypeSafe’s Jev 1.14 and OpenAI’s GPT-6 Luna, the only model behind OpenAI’s Decisions API. On Google’s own tests, Gemini Decision 1 tops seven of nine, but Jev wins two, including web-browsing decisions, and comes out cheapest. These are Google’s own figures and have not been independently checked. The comparison also makes the history plain: Google is not first to this idea. TypeSafe launched Jev in September, OpenAI unveiled its own Decisions API at DevDay on 29 September and opened a public beta on 6 October, and Microsoft followed with Microsoft-Decision-1, built on Alibaba’s Qwen, on 9 October. What Google is doing is joining a race that is already running — and betting that a cheaper, faster decision primitive is the piece that makes agents affordable at scale.

Why it matters. The cost of an agent is not the big, clever answer — it is the thousands of tiny judgements around it: is this tool call safe, is this document relevant, which model should handle this? Each one today costs a full generation. A model that prices the decision instead of the prose could cut that bill sharply, and at $0.10 per million tokens with no output charge, Google is pricing it like plumbing, not like a product. The catch is the one Google does not advertise: a score is only as good as the question, and a confident wrong answer is worse than a hesitant one. For developers, the sensible test is the one the site always recommends before trusting a new tool: run it on your own real tasks first, and compare it against the alternatives — Claude Fable 5.2, Gemini 4 Carbon and Muse Ember are the current yardsticks for what a model costs versus what it delivers.

Sources

  • Google
  • Gemini Decision 1
  • Gemini Decisions API
  • AI tools
  • AI agents
  • developers

Related articles

AI Tools

How AI Image and Video Generation Works

Type a sentence, get a picture — or a film clip. Here is the idea underneath modern AI image and video generators: diffusion, noise, and why video is the harder problem.

· image generation, video generation, diffusion, explainer