Google releases multimodal decision model Gemini Decision 1 and the Gemini Decisions API
On November 11, 2026, Google released the multimodal decision model Gemini Decision 1 and launched the Gemini Decisions API, which can be used through Google AI Studio. The API supports multimodal input of up to 256K tokens and up to 1,000 independent questions; latency is essentially constant when a single request contains 1 to 100 questions, and the median model computation latency is 50 to 100 milliseconds. Pricing is $0.10 per million input tokens, and output tokens are not billed; it leads on 7 of 9 benchmarks, and its calibration error is smaller than that of comparison solutions such as Gemini 3.5 Flash-Lite.
Why read this
AI score 75
The original text provides specific metrics such as latency, pricing, and calibration error, so readers can judge which workflows this type of decision model is suitable for.
Topics
Body · AI translation
On November 11, 2026, Google announced "Gemini Decision 1," a highly accurate and fast multimodal decision-making model, along with "Gemini Decisions API," a new API that makes it available through Google AI Studio.
Gemini Decisions API: Fast, calibrated multimodal decisions
Check out the API docs to learn more about getting started! https://t.co/T0Zylzg6xf— Logan Kilpatrick (@OfficialLoganK) November 11, 2026
Gemini Decision 1 is a multimodal model specialized for decision-making, similar to "TypeSafe Jev" and others. It evaluates options without generating a response token by token. In the company's measurements, with multimodal input and context of up to 16K tokens, the median model computation latency was 50 to 100 milliseconds, and end-to-end it was 150 to 250 milliseconds. It also reported that latency remained nearly constant when the number of questions per request ranged from 1 to 100.
The Gemini Decisions API built on this allows users to pass a single multimodal input of up to 256K tokens, including text, images, audio, video, and PDFs, along with up to 1,000 independent questions about that input. Gemini Decision 1 can directly process text, images, audio, video, and PDFs in a single forward pass, and it can also combine multiple modalities into one shared input, such as call recordings and account metadata, or insurance claim details scanned from a PDF and damage photos.
In the company's benchmarks for agent routing, reasoning, ranking, multimodal understanding, and other tasks, the combination of Gemini Decision 1 and the Gemini Decisions API recorded top performance in 7 of 9 benchmarks, including having the smallest calibration error—which represents the gap between predicted probabilities and actual accuracy rates—compared with Gemini 3.5 Flash-Lite using the Interactions API, TypeSafe Jev 1.14, and others.
Pricing for the Gemini Decisions API is $0.10 per 1 million input tokens, and output tokens are not billed.
The Gemini Decisions API allows checking by combining the following three types of questions.
yes/no judgment (noul): Estimates and returns the probability that a specified condition holds. A true/false explanation criterion can optionally be specified. Multiple choice (choice): Returns the result selected from 2 to 255 options and the normalized probability of each candidate. Graded evaluation (score): Scores along an ordered evaluation rubric and returns a continuous score and the probability of each grade.
As a result of these, a "confidence score" is output along with the probabilities. The confidence score (0 to 1) represents the probability that the model estimates that "the top answer is correct," allowing code to route ambiguous cases to fallback processing or have a human reviewer confirm them.
Example use of the Gemini Decisions API: With a photo of a counter and text describing the situation as a shared input, it can simultaneously evaluate "whether liquid has spilled," "which cleaning method is appropriate," and "how severe the soiling is." Each question evaluates the shared input independently and does not reference other questions or answers. What to do next by combining the results is determined by the application-side code.
For detailed usage of Gemini Decision 1 and the Gemini Decisions API, refer to their respective documentation.