LLM Playground

About this project
LLM Playground is a collection of interactive experiments exploring how AI models interact, persuade, and reason together. In the Debate Arena, two models argue opposing sides of a topic while an audience of diverse models tracks its shifting opinions in real time. The Compression Test challenges you to describe concepts in the fewest correct words, with a model judging accuracy and conciseness. The Evals page sends one question to many models in parallel and has a judge rate each response pass or fail, with cost surfaced as results stream in.
The model lineup spans 15 models across OpenAI, Anthropic, Google, DeepSeek, and others, all accessed through OpenRouter. The experiment list keeps growing: newer additions include an interactive concept explainer, a reverse Wordle where the model guesses your hidden word, and a game of 20 questions.