---
title: 'Gemini Just Went Absolutely WILD 🔥'
source: 'https://youtube.com/watch?v=B2xQKfbYWmA'
video_id: 'B2xQKfbYWmA'
date: 2026-11-17
duration_sec: 552
---

# Gemini Just Went Absolutely WILD 🔥

> Source: [Gemini Just Went Absolutely WILD 🔥](https://youtube.com/watch?v=B2xQKfbYWmA)

## Summary

Google has launched Gemini Decision One, a new AI model designed for rapid decision-making rather than text generation. It can process up to 1,000 questions about a single input in one request, returning answers with confidence scores in milliseconds. The video explains the three question types, real-world applications, and practical usage through Google AI Studio.

### Key Points

- **Gemini Decision One Launch** [00:17] — Google launched Gemini Decision One on November 11, 2026, powering the Gemini Decisions API. Unlike normal Gemini, it doesn't write; it makes choices.
- **How It Works** [00:45] — You provide one input (text, image, audio, video, or PDF) and ask up to 1,000 questions about it. The model returns an answer and confidence score for each question.
- **Speed** [01:10] — The model thinks in 50-100 milliseconds, with full round-trip latency of 150-250 milliseconds. However, these are Google's own test measurements, not a guarantee for all workloads.
- **Three Question Types** [01:24] — Type 1: Yes/No (called 'null' in Google's code). Type 2: Multiple choice (2-255 options). Type 3: Score (e.g., low, medium, high, critical). You can mix all three in one request.
- **Real-World Example** [02:18] — Using a customer support message, you can ask: Is this a login problem? Which team should handle it? How urgent is it? One request yields three decisions for routing and prioritization.
- **Google's Internal Use** [03:50] — Google uses the API in four places: Gemini Notebook (model selection, cutting wait by up to 40%), Google Stitch (Rapid Canvas design tool), Google Labs app CC (emoji reactions), and Google Search (safety and risk detection).
- **How to Use It** [04:44] — Access the Decisions Playground in Google AI Studio, paste text, set up questions, and run. You can export code for your own apps. Sample apps: Doorstep Parcels, Speed Sketch, Find the Zookeeper.
- **Developer Integration** [05:49] — Works with Python via the Google Gen AI package (version 3.2+). Set up API key and call the Decisions tool with model name 'Gemini Decision One'. Keep API key secret.
- **Caveats** [06:15] — Speed is not guaranteed for huge inputs or 1,000 questions. Input limit is 256,000 tokens. High confidence doesn't mean correct. It doesn't write normal answers. Beta only in developer API. Free tier may use data for product improvement.
- **Independent Benchmark** [07:21] — On the Artificial Analysis Decision Index, Gemini Decision One performs well but isn't number one. It's strong and fast but not the winner at everything.

### Conclusion

Gemini Decision One represents a shift from AI that talks better to AI that decides faster, which is crucial for AI agents that make countless small choices. By automating these decisions, businesses can save time and focus on higher-level thinking.

## Transcript

It doesn't chat. It decides. One request, up to 1,000 questions, answers inside its own products, and it's so easy even my coffee could set it up. it works and how to use it. So here's
the big news. On November 11th, 2026, Google launched something called Gemini Decision One. It powers a new tool called the Gemini Decisions API. And this is not like normal Gemini. Normal Gemini writes, you ask a question, it
doesn't do that. It doesn't write at all. It makes choices. Think about it talks for 10 minutes when you ask what time it is. Gemini Decision One is a fast, simple, done. Here's how it works. You give it one
thing to look at. That could be text, it could be a picture, it could be audio, video, or a PDF. Then you ask it questions about that one thing, and you can ask up to 1,000 questions in one single request. It looks at all of them
in one go and gives you back an answer for each one, plus how sure it is about each answer. Now why does that matter? Because speed. Google says the model itself does its thinking in about 50 to 100 milliseconds. And the full trip from
when you send it to when you get the answer back is around 150 to 250 That's faster than you can blink. But stick with me because there's a catch with those numbers and I'll tell you exactly what it is later in this video.
There are three types and this is the part that makes it really useful. Type one is yes or no. You ask a simple question and it tells you yes or no plus how likely that is. Fun fact, in Google's code they call this type null.
you might ask, does this message say something is broken? Yes or no. Great multiple choice. You give it a list of options, anywhere from two all the way up to 255 and it picks the best one and it shows you how likely each option is.
handle this, sales, support, or tech? This is perfect for sending work to the right place and it's huge for AI agents because an is always asking which tool should I use next. Type three is a score. You give it a scale like low,
medium, high, critical, and it tells you where the thing lands on that scale. So you might ask, how urgent is this? Perfect for ranking things and deciding what to do first. Now here's where it gets fun. You can mix all three in one
request. Let me give you a real example using the AI Profit Boardroom. Imagine can't log in and I can't find the coaching call replay. Please fix this today. With one request, you could ask three things. Yes or no, is this a login
should handle it? Access content or general help? And a score, how urgent is it from low to critical? One message goes in, three clear decisions come out, and your system knows exactly where to send it and how fast to move. That's a
shift. Google isn't just making AI better at talking, it's making AI faster at choosing what to do next. And think about how many places this fits. Yes or no questions are great for checking rules, spotting missing info,
Multiple choice is great for routing agents, handing out tasks, and picking the right tool. And scores are great for ranking leads, sorting support requests, and judging content, customer support, moderation, document checks, automation.
All of it runs on small choices, and this model is built just for that. Quick already using this inside its own products. If you're watching this and thinking, okay, but how do I actually use Gemini to sort my messages, route my
tasks, and stop doing all this by hand? That's exactly what we do inside the AI Profit Boardroom. It's the best place to scale your business with AI automation and save hours every week. You'll learn how to take tools like Gemini and turn
them into real workflows that run for you. There are step-by-step tutorials never stuck guessing. And with updates like this one dropping every week, now is the time to get ahead. The link is in the comments and description. Right back
developers. Google says it's already using it in four places. First, Gemini Notebook. Google uses the Decisions API to pick the right model and the right amount of thinking for each question. Easy question, use the fast path. Hard
question, think harder. Google says this cut the weight by up to 40% for everyday questions. Second, Google Stitch. That's Google's design tool. There's an experiment in it called Rapid Canvas where you change a design with your
voice or your touch and it uses this API to react super fast. Google even shows a video of it building a product flow at normal speed. Third, a Google Labs app called CC. It uses a model to pick emoji reactions from more than 200 options
without slowing the reply down. Fourth, and this one is big, Google Search. Google says it uses the API to help spot bad behavior and risky content involving MCP tools. So this isn't just for fun stuff, it's being used for safety too.
Google has something called the Decisions Playground inside Google AI Studio. You open it, sign in with your Google account, paste in your text, set up your questions, and hit run. You can test all three question types right
there. And here's the cool part. When you're happy with it, you can export the code and use it in your own app. Let me walk you through what that looks like with the boardroom example from earlier. You paste in the message. I can't log in
replay. Please fix this today. Then you add your three questions. Is this a login problem? Which team should handle it? How urgent is it? You hit run and how sure the model is. Now I'm not going to guess what those numbers will be. You
run it and see for yourself. But here's the key. Once you have those answers, to the right team, move it up the list if it's urgent, or flag it for a human to check if the model isn't sure. Google also made three sample apps you can play
with. They're called Doorstep Parcels, Speed Sketch, and Find the Zookeeper. They're all linked in Google's Launch Post so you can see how this works with pictures and real apps, not just text. And if you're a coder, it works with
Python. You install the Google Gen AI package, set up your API key, and call the Decisions tool with the model name Gemini Decision One. Google's guide says you need version 3.2 or newer of the package. One tip, keep your API key
secret. Never show it on screen and never share it. Now remember that catch I told you about. Here it is. Those super fast speeds are what Google measured on its own tests. They're not a promise for every job. If you stuff in a
really huge input or ask all 1,000 questions at once, it may not be that fast next. The 1,000 questions all look at the same one input, so it's not 1,000 different documents in one go. It's one thing looked at 1,000 ways. Next, the
thing looked at 1,000 ways. Next, the input can be big up to 256,000 tokens, the formats it supports. You can't just paste a YouTube link in and call it a day. Next, a high confidence score does not mean the answer is right. It means
the model feels sure. Those are not the same thing. So for important stuff, keep a human in the loop. Next, it does not write normal answers. That's the whole point. So if you need a blog post or an email, you still need a regular Gemini
support context caching or the batch API right now. Next, right now it's in the Gemini developer API in beta. Google says wider Google Cloud support is coming soon. And one more, if you use the free tier, Google says what you send
may be used to improve its products. So don't paste in private customer info is? Google says it leads on its own set of tests, but there's also an independent test called the Artificial Analysis Decision Index. And on that
one, Gemini Decision one does well, but it's not number one. Some other models score higher, so it's strong, it's fast, but it's not the winner at everything. So here's the big takeaway. Most AI news is about models that talk better. This
one is about AI that decides faster. And when you think about AI agents, that's huge. An agent has to make tiny choices all day long. Which tool? Which team? Is this urgent? Is this safe? If each of those choices takes a blink instead of
seconds, your whole system gets faster. The future of AI agents may depend just brains. Think about your own business for a second. How many tiny choices do you or your team make every day? Is this email important? Who should reply? Is
first? Most of those are simple calls. They just eat up time. That's exactly built for. You keep the big thinking, let the fast choices run on their own. Now, watching this video is a great start, but knowing about Gemini Decision
different things. Inside the AI Profit Boardroom, we help you go from AI news how to build simple automations that sort your messages, route your tasks to the right place, and flag what's urgent, just like the decision examples you saw
today. It's the best place to scale your business and save time with AI systems and start using them. The link is in the comments and description. I'd love to And if you want to keep learning for free, join the AI Success Lab. It's a
free AI community. You'll get the notes from this video so you don't have to decision types or the limits I covered. You'll also get step-by-step processes and SOPs you can follow, plus over 100 AI use cases just like the ones in this
video. It's the easiest way to keep learning AI and actually put it to work. description. And remember, Julian reads every comment, so drop yours below. Which decision would you automate first? See you in the next one.
