🤯 Gemini 3.5 "Snowbunny" Leaked: 3000 Lines of Code Emulators in One Prompt!

Introduction

If you watch AI news, things are changing very fast. But what I’ll share about Gemini 3.5 today (people call it "Snowbunny") is a big surprise. This is not a small update.

We are talking about an AI brain that can write over 3,000 lines of code at once, make music, and build a professional website in just a few minutes.

If you are a beginner or you know a lot about tech, this post will help you see why Gemini 3.5 is so important.

I. What Is Gemini 3.5 Snowbunny and Why Is It Special?

Gemini 3.5 Snowbunny (internal code DN9) is a powerful upgrade to Google's AI ecosystem currently in leaked testing. Here is a concise summary of its capabilities:

Key takeaways

• Leaked Name: Known as Snowbunny or DN9/D13 within the developer community.

• Multimodal: Handles text, coding, vector art (SVG), and native music simultaneously.

• Capacity: Can generate over 3,000 lines of code in one prompt without losing context.

• Goal: Designed to build entire professional products rather than just simple scripts.

First, let's talk about the name. Right now, Google has not officially sold a version called Gemini 3.5. What people are talking about is an early version that was leaked. Coders call it "Snowbunny" because they found codes like "DN9" or "D13" inside Google AI Studio.

You can think of Gemini 3 Flash as a very fast student. It does well on hard science tests for PhD levels (90.4% on GPQA Diamond). But Gemini 3.5 Snowbunny goes much further. It does not just answer questions. It is like a creative machine that can do many things.

The biggest change is how it handles different types of data. It can understand and make:

1. Websites and apps in one go.

2. Clear vector art (SVG).

3. Real music and sounds from your words.

4. Hard coding systems with thousands of lines.

It is cool to see an AI do so many things. But is Gemini 3.5 actually smarter than the older models? Let's look at the tests below.

II. Is Gemini 3.5 Really Smarter than the Older Versions?

Gemini 3.5 has superior logical depth, scoring 80-88% on the Hieroglyph benchmark, nearly double its competitors. It excels at identifying "trick" details in complex scenarios (68.5% accuracy) and solving entirely unfamiliar visual puzzles.

Key takeaways

• Hieroglyph Score: 80-88% (vs. GPT-5's 45%).

• Precision: Corrects subtle "trick" details missed by other models.

• Logic: Shifts from pattern matching to genuine lateral reasoning.

To know if an AI is smart, scientists use tests. With Gemini 3.5, two tests really surprised me. They show that this AI starts to think like a real person.

1. The Test for “Misguided Attention”

This is a fun test. People give the AI a trick question. If the AI only copies what it learned, it will fail.

For example, there is a famous problem about a train (Trolley Problem). A train is going to hit 5 people. Do you pull a lever to hit 1 person instead?

But in this test, they add a small detail: the 5 people are already dead. Most other AI models miss this and start talking about right and wrong like a robot.

But Gemini 3.5 noticed the detail. It gave the right answer. It got 68.5% on this test. This is better than many other top AI models. This means you can trust it to read your words very carefully.

2. Thinking Logically with the Hieroglyph Test

The Hieroglyph test is like a hard code-breaking game. It does not ask the AI for facts found in books or on the internet. Instead, it gives strange symbols and asks the AI to find the hidden rules.

Experts call this "lateral reasoning". This means the ability to think in new ways to solve a problem the AI has never seen before.

In this test, Gemini 3.5 really surprised me. It got a very high score between 80% and 88%. To help you see the big difference, look at this comparison chart:

Based on this image, we can see a huge gap between the models:

• The older Gemini 2.5 Pro only reached 20%.

• GPT-5-reasoning reached 45%.

• Meanwhile, Gemini 3.5 Snowbunny is at 80%+, which is much higher than all other models.

Now that we know Gemini 3.5 is great at solving puzzles, you will not be surprised to see it build even bigger things.

III. Build Professional Websites with Gemini 3.5

One thing I like most about Gemini 3.5 is building websites. You do not need to know HTML or CSS to start. It means you just describe the vibe or the look you want, and the AI does the hard work.

1. How the Work Happens

When I tried to make a website, it worked like this:

• Step 1: You describe your idea. For example: "Make a coffee shop website. Use a simple style and warm brown colors. Add a menu and a price list."

• Step 2: Gemini 3.5 writes all the code.

• Step 3: You ask for small changes, like adding a "dark mode" or making images move smoothly.

• Step 4: You just download it and use it.

2. A Detailed Prompt You Can Try

Here is an example you can copy and try on Google AI Studio:

Act like a professional website designer. I want you to make a website for a cake shop. The style should be fancy with cream and light gold colors.

Specific needs:

A top part (Hero section) with big text: "Cakes From The Heart."

A list of 3 cakes: Croissant, Strawberry Cake, and Chocolate Cake. Each needs a short description and a price.

A contact part with a map and phone number.

The website must look good on phones and computers.

Write all the code in one single HTML file. Include CSS and JavaScript too.

The result is a professional website in about 8 minutes. It costs about 0.38 USD. This is much cheaper than other AI models I have used.

IV. Gemini 3.5 Creates Original Music and Audio From Text

Can you believe it? Gemini 3.5 does not just write text. It also knows how to make music. This is a big step. Usually, we have to use other apps like Suno or Udio.

1. The Secret Test Revealed by the Community

This feature is not public yet, but we have strong proof that it exists. A user named Legit spotted a new Gemini model being "A/B tested" over the weekend on Google AI Studio.

Please look at this leaked post:

According to this shared image, the new model is already working and it "outperforms" other music tools. It does not just give you notes on a paper; it generates a real audio file that plays right inside the chat.

This confirms that Google is ready to bring native music creation to everyone very soon.

2. Why Is This a Game-Changer for You?

The real magic is that Gemini 3.5 understands your entire project. Because it is a "multimodal" model, you can do everything in one single chat.

Imagine this future workflow: You ask the AI to write a funny video script first. Then, instead of switching to another app, you simply say: "Now, compose a background track that matches the mood of this script."

You won't need to describe the "vibe" again because the AI already knows your story. This saves you a lot of energy because you don't have to explain your idea twice.

V. Design Professional Vector Graphics with Gemini 3.5

Now that you have your audio, you need something for people to look at. That is where vector graphics (SVGs) come in. Unlike normal photos that get blurry when you zoom in, SVGs stay sharp and clear at any size.

Gemini 3.5 is surprisingly good at understanding how to draw these shapes using code. It acts like a professional designer who can turn your words into clean visual icons.

1. The "Cyberpunk Robot" Leak

A user shared a very interesting test on social media. They showed a collection of robot images created by a "new Gemini Pro model" during a secret A/B test on Google AI Studio.

Looking at the image they shared, the results are amazing because of the consistency:

• Matching Theme: All robots use the same neon blue and purple colors, making them look like they belong to the same team.

• Unique Designs: Even though they share a style, each robot has a different shape and look.

• Clean Code: The user confirmed these are SVGs, meaning the code is neat enough to use in professional design tools.

2. The “Pelican on a Bicycle” Challenge

Asking for a "pelican riding a bicycle" has become a standard test for "vector intelligence." It is very hard to get right, and most models just draw broken shapes.

Simon Willison performed this exact test to compare the models. While we cannot claim that Gemini passes this test perfectly every time, reports on the "Snowbunny" version show a clear improvement.

According to the analysis, the model handles three things much better now:

• Unusual Combinations: It successfully mixes weird objects like a bird and a bike together.

• Correct Structure: The SVG code is built correctly without errors.

• Spatial Logic: The bird is actually sitting on the bike, not floating in the air nearby.

This is basically asking: "Does the AI really understand 2D geometry, or is it just copying patterns?" For Gemini 3.5, the answer seems to be that it is at least starting to really grasp these concepts.

VI. Gemini 3.5 Can Code Entire Complex Systems by Itself

We are used to AI writing small scripts, but the Snowbunny tests show something much bigger. A user asked Gemini 3.5 to build a full Nintendo Game Boy emulator, and the result was shocking.

1. 3,000 Lines of Code in One Prompt

Coding an emulator is incredibly hard because you have to mimic hardware. In this test, the model generated over 3,000 lines of code in just one single run. The resulting project could actually launch real game ROMs with only a few tiny manual fixes.

Developers are already proving this is real:

• Jerrod Lew shared a working demo that got the coding community very excited, a complete game built entirely inside AI Studio.

• Chetaslua followed up with similar results, successfully running complex game simulations that usually take humans days to build.

The result was a system that could actually play games. It needed a few small fixes from a human, but seeing an AI understand such a big and hard structure is something we have never seen before.

2. The Future of Coding

This is not just about solving simple coding puzzles (like LeetCode). It proves that Gemini 3.5 can architect an entire system. It understands how to connect different complex parts like:

• CPU emulation and memory management.

• Input handling and timing logic.

• Display pipelines to show graphics.

It signals that Gemini is finally ready for system-level code generation, allowing you to build full products or complex tools just by describing them.

VII. How to Try the Leaked Gemini 3.5 in AI Studio

After hearing about these cool things, you probably want to try it. But since this is a test version, you might not see it right away.

1. Use Google AI Studio

The best way now is to use Google AI Studio. This is where Google lets people try the newest tech. Follow these steps:

• Go to the Google AI Studio website.

• Log in with your Google account.

• In the part to pick a model, look for Gemini 3 Flash.

• Sometimes, you will be lucky and see codes like DN9. That is when you are using Gemini 3.5.

2. Things to Remember

Because this is not the final version:

• It can change any time. It might make music well this morning, but Google might turn it off later to fix things.

• Do not worry if you do not see the "Snowbunny" code. Even the regular Gemini 3 Flash is very strong for daily work.

VIII. Save Time and Money Using Gemini 3.5 for Automation

Good news! Google's price for Gemini 3.5 and Flash is very low. It is cheaper than many other AI models.

A token is like a small piece of a word. One million tokens is a lot of data, like many books. Making a website or a song only costs a few cents. This price is good for students or people starting a new business.

IX. Tips for You to Use Gemini 3.5 Better

The most important thing is how you control it. To do well with Gemini 3.5, remember these 3 rules:

1. Be very specific: Do not just say "make a website." Describe the colors, why you need it, and who will look at it.

2. Describe the vibe: Do not be afraid to use words for feelings like "warm," "modern," or "simple." The AI understands these words now.

3. Work together: Do not let the AI do 100% of the work. Let it do 80%, and you spend time to fix the last 20%. This is the best way to get a perfect result.

The world is changing. We used to just chat with AI, but now we build things with AI. Gemini 3.5 Snowbunny is the best example. I hope this post helped you see clearly and feel excited about the future.

Conclusion

We have walked through a lot of exciting tests today. From the "Misguided Attention" riddle to the "Pelican on a Bicycle" challenge, and the amazing Game Boy emulator, one thing is clear: Gemini 3.5 "Snowbunny" is a beast.

With this new model, you are no longer just a writer or a coder. You are a Director. The gap between "having an idea" and "making it real" has never been smaller.

You do not need to wait for the official release to start learning. The future is already here in these test versions.

If you are interested in other topics and how AI is transforming different aspects of our lives or even in making money using AI with more detailed, step-by-step guidance, you can find our other articles here:

• Building Apps with Bolt: A No-Code Guide to Turning Ideas into Reality

• 10 AI Tools Giving Developers The Power Of A Full Dev Team

• Detailed Guide: How To Automatically Get Unlimited High-Quality LinkedIn Jobs*

• Prompt Engineering Automation: Build a Mini AI Assistant with n8n

• Discover My Ultimate AI Tools Productivity Kit for 2024*

  *indicates a premium content, if any