What changed
Nintendo’s classic SNES console now runs a chatbot powered by a 2.2-million-parameter language model, a distilled version of OpenAI’s GPT-OSS-120b. The system uses the Super FX 2 chip, a coprocessor originally designed to enhance SNES games with faster 2D and 3D graphics. This chip, running at 21.47 MHz, accelerates the chatbot’s responsiveness, allowing users to interact with a version of ChatGPT on a retro console.
What this means in practice
The SNES app, developed by Vintage Computing blogger Benj Edwards, can be run via an emulator or on a real SNES with a flash cart that supports Super FX 2 and MSU-1 chips. Users can choose between a 4MB model or a 19MB version that includes 150,000 Simple English Wikipedia articles stored in the cartridge’s MSU-1 data file. The chatbot can respond to text input via a keyboard, with sound effects and a chip-tune soundtrack enhancing the experience. However, the model’s limited size and knowledge base result in responses that are often quirky or incomplete.
Limitations and availability
The chatbot’s usefulness is constrained by its small size, which is millions of times smaller than the models behind ChatGPT. The model’s knowledge base is limited to the integrated Wikipedia content, and it cannot perform complex tasks like advanced calculations. The SNES app is available through an emulator or via a real SNES with compatible hardware. Users must have a flash cart with Super FX 2 support and MSU-1 compatibility for the enhanced version. The project is primarily a novelty, offering limited practical value beyond its retro appeal.
