Video tutorial · The Metaverse Guy
Gemini 3 Pro Code Review: Testing on a Real Next.js Project
I used Gemini 3 Pro as a coding agent in VS Code to build authentication, a chat interface, a landing page, and a 3D Rubik's cube simulator – here's how it compared to Claude Sonnet 4.5.
- Published
- 18 November 2025
- Length
- 15 min
- Views
- 6,445 views
I put Google's Gemini 3 Pro through a series of real-world coding challenges using Kilo Code in an existing Next.js project. It handled Clerk authentication, a full-stack AI chat interface, a modern landing page, and even a 3D Rubik's cube simulator that defeated other top models. Here's the full breakdown of its performance and where it now sits in my AI coding priority list.
Introduction & Setup
Gemini 3 Pro has arrived with Google's promise of a massive 1 million token context window and state-of-the-art reasoning. I wanted to see whether it could dethrone the current coding champion, Claude Sonnet 4.5, so I hooked it up as an AI agent inside VS Code using Kilo Code and pointed it at an existing Next.js project.
I planned four real-world, multi-step tasks that would push its understanding, planning, and design abilities – the kind of work you actually do as a developer.
Task 1: Clerk Authentication Integration
The first task was to integrate Clerk into the application for sign-in and sign-up. I gave the instruction and Gemini3 Pro immediately produced a step-by-step plan – it knew exactly which files to create, how to configure the middleware, and how to wrap the app.
It generated all the necessary code and I was able to test the full Clerk sign-in flow without any manual fixes. The plan was accurate and the implementation worked on the first run.
Task 2: Full-Stack AI Chat Interface
Next, I asked it to build a complete AI chat interface that would use the Gemini 3 Pro model itself via OpenRouter. This meant creating a new API route, handling streaming responses, and building a clean chat UI.
Gemini3 Pro created the API route and the front-end components, and after I plugged in my OpenRouter key, I had a working chatbot inside the project. It even suggested proper error handling and loading states.
Task 3: Front-end Landing Page Design
I then tested its design sense by asking for a modern, professional landing page. I wanted to see if it could produce something that looked polished without any design hand-holding.
The generated page had a clean layout, good use of spacing, and a contemporary feel. I evaluated the design quality and was genuinely impressed – it felt like something you'd ship to production with only minor styling tweaks.
Task 4: The 3D Rubik's Cube Simulator Challenge
The ultimate test was a challenge I've given to many top models: build a fully functional 3D Rubik's cube simulator. This requires complex 3D maths, proper cube state management, and interactive controls – models like Claude 4 and GPT-5 have all failed here.
Gemini 3 Pro generated the complete code, and when I ran the simulation it worked flawlessly. The cube rendered correctly, the rotation logic was sound, and I could interact with it. This was the moment I knew it wasn't just hype.
Final Verdict & Cost
After running all four tasks, I looked at the overall results and the pricing picture. The 1 million token context window is a real advantage for big projects, and across these multi-step engineering tests it matched or beat Claude Sonnet 4.5.
The Rubik's cube simulation, in particular, showed a level of 3D reasoning I hadn't seen before from an AI coder. For now, Gemini 3 Pro has moved to the top of my AI coding priority list.
Tools used in this video
- Gemini 3 Pro
- Kilo Code
- VS Code
- Next.js
- Clerk
- OpenRouter
- Claude Sonnet 4.5
- GPT-5
Questions
- What is the context window of Gemini 3 Pro?
- Gemini 3 Pro has a 1 million token context window, which allows it to process very large codebases and retain massive amounts of context in a single session.
- How does Gemini 3 Pro compare to Claude Sonnet 4.5 for coding?
- In my tests it handled everything Claude could do, plus succeeded on a 3D Rubik's cube simulator that Claude 4 could not build, making it a very strong competitor.
- Can I use Gemini 3 Pro as an AI coding agent?
- Yes, you can pair it with tools like Kilo Code in VS Code and give it direct read/write access to your project files, allowing it to plan and implement features autonomously.
- What made the Rubik's cube simulator so difficult for other models?
- It requires handling complex 3D rotations, state management of the cube pieces, and a working interactive user interface – a combination that tripped up Claude 4 and GPT-5 completely.
Chapters
- 00:00 ,Introduction & Gemini 3 Pro Context Window
- 00:40 ,Price & Expectations (vs. Claude Sonnet 4.5)
- 00:50 ,Setting up the Kilo Code Agent (Next.js Project)
- 01:05 ,Task 1: Next.js + Clerk Authentication Integration
- 02:00 ,Gemini 3 Pro's Step-by-Step Plan
- 04:00 ,Task 1 Complete: Testing the Clerk Sign-in
- 04:55 ,Task 2: Full-Stack AI Chat Interface Creation
- 05:15 ,Integrating Gemini 3 Pro into Our Project (via OpenRouter)
- 06:10 ,Gemini 3 Pro Creates API Route & UI
- 07:35 ,Task 2 Complete: Testing the Gemini 3 Pro Chatbot
- 08:30 ,Task 3: Front-end Landing Page Design Test
- 09:30 ,Evaluating the Modern Landing Page Design Quality
- 10:52 ,Task 4: The Ultimate Challenge - 3D Rubik's Cube Simulator
- 11:30 ,Why The Rubik's Cube Test Fails Most Models
- 12:48 ,Code Generation Complete (Running the Simulation)
- 13:30 ,Mind-Blowing Result: Flawless 3D Simulator
- 14:35 ,"Final Thoughts, Cost Summary & Competition Comparison"
- 15:10 ,Conclusion & My New AI Coding Priority List
Video description and links
This article was written from the video's description with the help of AI. Watch the video for the full walkthrough.



