Google Releases Gemini Decision 1: AI Answers Yes/No and Checks 100 Photos in 4.5 Seconds
Google has released Gemini Decision 1, an AI model that doesn't write text but gives short answers to questions about a file: yes/no, a choice from a list, or a rating on a scale. For example, given a photo of a parcel, it says whether the box is dented. In Google's demo, the model went through 100 images in 4.5 seconds, answering 16 questions for each, or roughly 1,600 answers. For sorting and checks, this format is easier to build into a pipeline than parsing free-form text.
AI-processed from RB.RU; edited by Hamidun News
Picture a warehouse where every parcel is photographed on a conveyor belt, and you need to tell whether the box is dented and whether any item has fallen out. A standard language model would answer with a paragraph of reasoning, from which the code has to extract the word "yes" or "no". Google has released Gemini Decision 1 to cut out that hassle: the model doesn't chat or write text, and instead returns a short decision right away. In the company's demonstration, it checked 100 parcel photos in 4.5 seconds and answered 16 questions for each, producing roughly 1,600 answers. The questions included box damage and items that had fallen out of the packaging.
How the model answers
According to Google, Gemini Decision 1 is not designed for dialogue or text generation. Its job is to quickly analyze a file and answer a direct question. It works with text, images, audio, video, and PDF documents.
There are three answer formats:
- "Yes" or "No": a choice between two options, for example whether the packaging shows any damage.
- "Choose": one option from a list of two to 255 items.
- "Rate": an assessment of an object on a predefined scale.
The model attaches a confidence score from 0 to 1 to each answer. The closer the value is to one, the more confident the system is. In practice, this is a useful lever: high-confidence answers can be accepted automatically, while doubtful ones go to a human reviewer.
Where the model is already used
Google says it already uses Gemini Decision 1 in its own services. Internally, the model determines the type and complexity of a user's request and routes the task to the appropriate AI model. The logic is clear: simple questions don't need to run through a heavy and expensive model.
The second example is Google Search. There, Gemini Decision 1 helps detect malicious commands and attempted attacks via MCP tools. MCP is the protocol through which AI agents connect to external services. For anyone building agents, this is a telling example: the question "is this command dangerous or not" fits neatly into a yes/no format.
How to connect
Alongside the model, the company released the Gemini Decisions API. Developers use it to connect the model to their own applications and AI agents. Access is available through the Gemini API and Google AI Studio, and connecting requires an API key.
According to Google's description, the model suits tasks such as:
- classifying data in third-party software;
- picking the right option when a user is uncertain;
- scoring results against defined criteria;
- checking incoming files: photos, audio, video, PDFs.
A model with a strict answer format has a practical advantage: the output is easier to validate and measure, and the code doesn't break because of extra words.
What is still unknown
All speed figures come from Google's own example. The source material contains no data on accuracy, pricing, or independent testing, and no comparison with standard Gemini models on the same tasks. It is also unclear how closely the confidence score matches the actual share of correct answers.
Therefore, before deploying the model, test it on your own data. Take a labeled set, for example photos with known damage, and see how often "yes" turns out to be correct. Also determine a confidence threshold at which the error rate stays acceptably low.
If the testing holds up, you get a compact component for file-processing pipelines. It answers simple questions quickly, while expensive general-purpose models remain for complex tasks.
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026MIRA Neural Network Renders Rocket League at 20 FPS, but Ping Hits 460 ms
- Nov 21, 2026Google built a closed copy of the internet with fake banks to train Chrome's AI agents
- Nov 21, 2026MLPerf Training Adds Nvidia Rubin and TPU Ironwood in First Appearance: 153 Systems
Key Questions Answered
How the model answers?
According to Google, Gemini Decision 1 is not designed for dialogue or text generation. Its job is to quickly analyze a file and answer a direct question. It works with text, images, audio, video, and PDF documents. There are three answer formats: "Yes" or "No": a choice between two options, for example whether the packaging shows any damage. "Choose": one option from a list of two to 255 items. "Rate": an assessment of an object on a predefined scale.
Where the model is already used?
Google says it already uses Gemini Decision 1 in its own services. Internally, the model determines the type and complexity of a user's request and routes the task to the appropriate AI model. The logic is clear: simple questions don't need to run through a heavy and expensive model. The second example is Google Search. There, Gemini Decision 1 helps detect malicious commands and attempted attacks via MCP tools. MCP is the protocol through which AI agents connect to external services.
How to connect?
Alongside the model, the company released the Gemini Decisions API. Developers use it to connect the model to their own applications and AI agents. Access is available through the Gemini API and Google AI Studio, and connecting requires an API key. According to Google's description, the model suits tasks such as: classifying data in third-party software; picking the right option when a user is uncertain; scoring results against defined criteria; checking incoming files: photos, audio, video, PDFs.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.