ChatGPT vs DeepSeek AI-generated code is evolving rapidly, but which model writes better C++ for game development? In this experiment, I put ChatGPT o3-mini-high and DeepSeek to the test with the same prompt: create a C++ game development project using SFML. Both AI models attempted to generate an infinite platformer where the player can move, jump, and defeat enemies.
I walked through ,reviewing their coding approaches, debugging issues, and configuring the development environment in VSCode with vcpkg. While ChatGPT o3 provided a more structured implementation with smoother gameplay, DeepSeek struggled with infinite level generation and enemy behavior.
Write a simple 2D platform-style game using SFML, where the player controls a square that can move left, right, and jump, similar to Mario. The level should be procedurally generated and infinite, with the game recording the maximum distance the player travels forward.
Both AIs tackled the challenge differently, so let’s break it down and compare their results.
📺 Watch the Full Video
🎥 YouTube Video: Watch Here
🖥️ AI-Generated Code
ChatGPT’s Approach
ChatGPT provided a well-structured C++ implementation using SFML, with clear logic for:
- Player movement, gravity, and jumping
- Procedural level generation for an infinite world
- Collision handling for platforms and enemies
- Proper enemy defeat mechanics by jumping on their heads
The result? A playable and smooth experience with a structured game loop.
DeepSeek’s Approach
DeepSeek also generated a working platformer but had some key issues:
- The level was not infinite—it failed to generate new platforms correctly.
- Enemies floated instead of walking on platforms.
- Jumping physics felt less refined and caused inconsistencies.
- The game reset improperly after falling, not tracking distance correctly.
Both AIs showed an impressive understanding of SFML and C++, but ChatGPT’s result was more polished and functional out of the box.
⚙️ Setting Up the Project
I used VSCode with vcpkg to set up the project:
- Created a new folder:
GPTVsDeepSeek - Initialized a CMake project
- Configured vcpkg for SFML dependencies
- Installed SFML and updated CMakeLists.txt
- Built both AI-generated codes for testing
If you want a detailed setup guide, check out my full VSCode + SFML setup tutorial.
🏆 Final Verdict: Which AI Did Better?
After running both AI-generated projects, ChatGPT vs DeepSeek had a clear winner. ChatGPT o3 provided a more polished and playable experience, while DeepSeek struggled with several critical issues.
✅ ChatGPT o3’s Strengths:
- Smooth Gameplay & Physics: The movement felt fluid, with proper gravity, jumping mechanics, and responsive controls.
- Proper Procedural Level Generation: Platforms spawned correctly, allowing infinite gameplay as expected in a game development environment.
- Accurate Collision Handling: Jumping on enemies successfully eliminated them, and the player interacted with platforms as expected.
❌ DeepSeek’s Issues:
- Buggy Enemy Movement: Enemies often floated in mid-air instead of walking on platforms.
- Broken Level Generation: The game failed to generate an infinite world, meaning the player would eventually run out of platforms.
- Faulty Distance Tracking: The score counter didn’t function correctly due to the lack of proper level generation.
Final Thoughts
ChatGPT o3 outperformed DeepSeek, delivering a more structured and polished experience. While both AIs showed a solid understanding of game development, ChatGPT o3’s implementation felt significantly more refined.
🚀 Winner: ChatGPT o3-mini-high!
📂 Get the Full Source Code
🔗 GitHub Repository: View the Code
You can download and test both AI-generated projects for yourself!
🔥 What’s Next?
Should I test these AIs with Raylib or SDL next? Let me know in the comments!
For more AI coding experiments, subscribe to my YouTube channel: Rambod Dev.
February 2025 – AI Coding Evolution Continues 🚀