Google launches Gemini Decision 1 — an AI that scores answers without writing them out

What you could do with it

• A developer building an app could feed up to 1,000 yes-or-no or multiple-choice questions about text, images, audio, video, or PDFs into the API and get scored probability answers back in under a quarter of a second ².
• A business operator could use the model to automatically route customer queries to the right tool or team by scoring which option best fits each incoming request ².
• A content moderator could run a policy-check batch across thousands of documents or videos at once, receiving confidence scores for rule violations without waiting for long AI-generated explanations ².

What Gemini Decision 1 does differently

On 11 November 2026 Google announced the Gemini Decisions API, powered by a new model called Gemini Decision 1 (²). Rather than generating answers token by token like most generative AI tools, it scores candidate options in a single forward pass and returns probability numbers .

Users can submit up to 1,000 independent yes/no, multiple-choice or rubric-scoring questions in one go, fed a multimodal input that may include text, images, audio, video or PDFs with a context window of up to 256K tokens ². End-to-end latency lands between 150 ms and 250 ms, and the median model compute sits at just 50–100 ms . The cost is fixed at $0.10 per million input tokens, with output tokens billed at zero ¹.

For comparison, the blog Simon Willison noted that competing Jev is cheaper at 4.2 cents per million tokens but only handles text and JSON, while OpenAI's GPT-6 Luna decision model shares the same pricing as Gemini but processes text and images ¹. Gemini Decision 1 covers every modality — text, images, audio, video and PDFs .

How confident the answers are

An early test ran four questions against a robot-stomping video: whether it contained a dragon, a robot, was set in France, or was set in the USA ¹. Gemini returned probability scores — 0.049 for dragon, 0.923 for robot, 0.083 for France, 0.923 for the USA — rather than hard yes-or-no values . Simon Willison observed that Google appears deliberately calibrated: unlike Jev and OpenAI, which return clean 1.0 or 0.0 outputs for binary questions, Gemini keeps probabilities slightly softer to avoid overconfidence .

Google states that Gemini Decision 1 leads competitors on benchmarks spanning agent routing, reasoning, ranking and multimodal understanding, with the lowest calibration error across the suite ². Performance remains nearly flat whether you ask one question or a hundred .

Who can use it now

The API is available immediately through Google AI Studio with Python, JavaScript and REST documentation provided ². Min Choi, product manager at Google DeepMind, and Felix Yu, principal scientist at Google DeepMind, introduced the tool publicly . For anyone building apps that need quick, accurate decision-making at scale, this could cut both costs and wait times dramatically.

This story reports on a trend for information only. It is not advice or encouragement to do any of it.

How good is it

Google tested Gemini Decision 1 on benchmarks spanning agent routing, reasoning, ranking, and multimodal understanding, and says it leads the suite compared with rivals Jev and OpenAI's GPT-6 Luna decision model ². The key measure Google highlights is calibration error — how much the model's confidence score drifts from reality — and they say Gemini Decision 1 has the lowest calibration error of the competitors . In practice that means when the model says "I'm 90% sure," it turns out right about 90% of the time ¹. A real-world test showed the model giving soft probabilities like 0.923 for "robot" and 0.049 for "dragon" in a video, whereas Jev and OpenAI return clean 1.0 or 0.0 answers .

Key facts

• Gemini Decision 1 costs $0.10 per million input tokens with free output
• End-to-end latency is 150–250 milliseconds
• Supports up to 1,000 questions across text, images, audio, video and PDFs
• Compared against Jev and OpenAI's GPT-6 Luna in independent testing
• Available now via Google AI Studio with Python, JS and REST guides

Sources

1. Gemini Decisions API
2. Gemini Decisions API: Fast, calibrated multimodal decisions