AI vs the World Cup
Five frontier AI models — Claude (Anthropic), GPT (OpenAI, the model behind ChatGPT), Gemini (Google), Grok (xAI) and DeepSeek — each predicted the entire 2026 FIFA World Cup before a ball was kicked: every group match with win/draw/loss probabilities, a full knockout bracket, and tournament awards. The complete set of picks was locked and hashed with SHA-256 (commitment published on the methodology page), so nothing can be edited after the fact. As real results come in, every prediction is graded with the Brier score, which rewards honest, well-calibrated confidence rather than lucky guessing, and the models are ranked on a live leaderboard.
What you can find here
- A live Brier-score leaderboard ranking the five models on graded results
- Every model’s match-by-match predictions, committed before the tournament and fresh live forecasts before each game
- Each model’s full predicted knockout bracket next to the real tournament bracket
- Predicted awards: champion, top scorer, Golden Ball and more
Final results (tournament ended 19 July 2026)
Spain won the World Cup, beating Argentina 1–0 in the final — a result all five models predicted. On the scoring side, DeepSeek finished with the best overall average Brier score (0.464), edging Google's Gemini (0.465) by the narrowest margin of the tournament; Gemini won the committed pre-tournament board, DeepSeek won the live board, and xAI's Grok picked the most match winners (120 of 176) but lost the calibration race. GPT (ChatGPT) finished last on all three boards.
The interactive leaderboard and match pages need JavaScript, but the full project is readable without it:
Match-by-match AI predictions · About · Methodology · FAQ · The pre-tournament picks · What the Brier score measures · Why Claude picked Argentina · Why Grok bet on Messi · GPT vs Gemini brackets · DeepSeek’s calibration · Where all five models agreed