Developer Uses Claude AI to Bring Super Mario 64 and Its Authentic Physics to the Game Boy Advance

The landscape of video game preservation and reverse engineering is undergoing a fascinating transformation, driven largely by rapid advancements in AI-assisted code decompilation and porting. In recent weeks, the gaming community has witnessed an unprecedented wave of ambitious projects. These efforts range from high-end modern achievements, such as getting modern blockbusters to run natively on advanced silicon like Apple’s M4 chip, to remarkable feats of technical ingenuity aimed at the opposite end of the spectrum. While pushing modern games onto futuristic hardware continues to grab headlines, a parallel movement is exploring what happens when developers attempt the reverse: taking beloved 3D classics and porting them backward onto aging, severely constrained retro hardware.

The latest example of this retro-focused ingenuity comes from a developer who utilized Anthropic’s Claude AI assistant to bring Super Mario 64 to the Game Boy Advance. Shared by user Nervous_Cold8493 on the ClaudeAI subreddit, the project is capturing attention not merely because it runs on Nintendo’s classic 32-bit handheld, but because it meticulously preserves the exact movement, momentum, and collision mechanics that made the original Nintendo 64 release a timeless masterpiece.

This Super Mario 64 de-make preserves the titular character’s momentum

Over the years, hobbyists and independent developers have conceptualized and built numerous "demakes" of Super Mario 64 for older generations of consoles. However, many of these projects rely on simplified approximations, custom engines, or superficial recreations that manage to capture the visual aesthetic of the game while altering how the iconic platformer actually feels to play. Nervous_Cold8493’s approach stands apart because it focuses on a faithful recreation of the original source code’s underlying logic rather than a mere surface-level rewrite.

This distinction is crucial for a title where movement is arguably the core gameplay mechanic. In Super Mario 64, mastering Mario’s intricate physics system—from running and diving to executing precise long jumps and wall kicks—defines the entire experience. Losing that signature momentum would fundamentally break the feel of the game. By leveraging AI assistance, the developer managed to translate the complex behavioral code of a powerful 64-bit console into something manageable for a much simpler, highly restricted platform.

Outlining the technical reality of the project, the creator noted the severe limitations under which the code must operate. The demake’s movement and collision code runs directly under strict Game Boy Advance constraints, which means navigating a 16.8 MHz ARM7 processor completely devoid of a Floating Point Unit (FPU). Despite these hardware bottlenecks, the project currently exists as a fully playable demonstration featuring four distinct levels, rudimentary software 3D rendering, and Mario animated using the original game’s authentic animation data.

Furthermore, the implementation includes an optional, libsm64-style import mechanism designed to respect copyright boundaries. Instead of bundling proprietary Nintendo assets within the distribution repository, the software is engineered to read a user’s legally acquired ROM to extract Mario’s actual 3D model and the real level meshes. The repository itself contains no proprietary material originating from Nintendo, ensuring that the project remains centered on the technical achievement of the code translation.

Current demonstrations of the project have been successfully tested using the popular mGBA emulator, meaning the software has not yet been rigorously benchmarked or verified on physical Game Boy Advance hardware. Successfully bridging the gap between an emulator environment and actual retro silicon would represent a monumental milestone for the project, elevating it from an impressive software curiosity to a genuine hardware marvel. For enthusiasts interested in examining the technical implementation, reviewing the source code, or experimenting with the demo, the project is publicly accessible via its GitHub repository, offering a fascinating glimpse into what is possible when modern artificial intelligence meets classic console architecture.