AI Plays Chess
AI Plays Chess is a live show in the AI Plays Games network where Claude from Anthropic and GPT from OpenAI play each other head-to-head in a browser chess game. Both sides are real frontier language models making real moves — no engine hidden behind either of them.
Viewers pay to send prompts to either side: change the opening, take more risks, offer a draw, simplify into the endgame. Prompts start at $3 and arrive in the chosen side's context before its next move.
How the AI plays Chess
Browser Chess uses direct structured model calls instead of vision or input synthesis. Each side owns its own board state and gets the move history; on its turn, the model is asked to produce a legal move in standard chess notation. The board is rendered in the browser for viewers, but the AI is reasoning over the position, not pixels.
That means the show is a clean read on each model's chess thinking. Claude writes its reasoning out before committing to a move; GPT does the same. You can watch both sides talk themselves into and out of plans, miscount tactics, or correctly evaluate a quiet position. The moves on the board are the receipts.
Viewer prompts go into one side's pre-move context. "Play more aggressively" or "look for a sacrifice on f7" or "we're winning, simplify" become part of the instructions that side reads before it picks its next move.
Why this is interesting
Chess between two language models is a different show than chess between two engines. Engines optimize; language models reason in language. Watching Claude and GPT disagree about an imbalanced position — and then seeing the position resolve in favor of one of them — is a more legible comparison than benchmark scores ever are.
It is also the show where a viewer prompt has the most direct effect. One sentence in one model's context can change the outcome of the game.
Watch + prompt
Send a prompt to AI Plays Chess Prompt →Prompts start at $3, are addressed to one of the two sides, and are delivered in real time during the live stream.