AIWC26 — AI vs AI World Cup 2026 sweepstake

2 min read Original article ↗

Several large language models were handed an identical brief and asked to predict every group-stage scoreline before kick-off; that round is done. They're now predicting the knockout rounds as the bracket resolves, currently the Semi-finals. Their answers are published here unedited and scored against the real results. None of them have seen a football.

3 pts exact 90-min score · 2 pts correct winner · 0 for the wrong winner — knockout matches can't draw

Ranking Bot — method

Deterministic baseline. FIFA World Ranking of 1 Apr 2026; teams within 10 places of each other draw 1-1, otherwise the higher-ranked team wins 2-1. No model, no judgement. For the winner pick, a drawn scoreline goes to the higher-ranked side. (generated 2026-06-10)

Claude Fable 5 — method

Weighted quarter-final and knockout performances most heavily as the best signal of current form, with group-stage dominance and historical big-match pedigree as tiebreakers. Favoured teams winning comfortably in 90 minutes over those needing late goals or penalties. (generated 2026-07-14)

Claude Sonnet 4.6 — method

Weighted recent tournament form heavily, with quarter-final and R16 performances as primary signal. Also considered historical knockout pedigree and squad depth at this stage of the competition. (generated 2026-07-14)

GPT-5.5 — method

I weighted the knockout-stage performances most heavily, with group-stage form as supporting evidence. I also considered historical knockout pedigree, squad depth, likely fatigue after six matches, and the tendency for semi-finals between elite teams to be low-scoring. (generated 2026-07-14)

Gemini 3.1 Pro — method

Analyzed knockout-round defensive stability and offensive output, weighting France and Spain's strong defensive records against England and Argentina's higher-scoring, higher-conceding paths to the semi-finals. (generated 2026-07-14)

Gemini 3.5 Flash — method

Analyzed team momentum and defensive records from prior rounds, heavily weighting Spain and Argentina's consistent scoring efficiency and defensive resilience in high-stakes knockout matches. (generated 2026-07-14)

Human 1.0 — method

Picks made after seeing the quarter-final results; no separate prediction made for extra time or penalties, just the 90-minute winner and score. (generated 2026-07-14)