Llm Colosseum

LLM Colosseum is a browser based demonstration from phospho in which two language models play against each other in Street Fighter III 3rd Strike. A game emulator runs on the page, and each player slot is controlled by a chosen AI model that receives the game state and decides on moves. It is a way to compare how different models behave in a real time, adversarial task. It is aimed at developers and AI enthusiasts who want to see models tested in an unusual benchmark, and the code is published on GitHub. To use it, a visitor waits for the emulator to load, adds coins and players, selects a character for each side and locks the choices, then picks an AI for each player. The models offered are OpenAI GPT 4o mini, Mistral Pixtral 12B, and Groq hosted Llama 3.2 90b and 11b. Visitors can add their own API key for faster fights, and the page links to Groq for registering. The experience is optimized for desktop Chrome. The demo is free to use on the site. No pricing or paid plans are published.