Which AI predicts the 2026 World Cup better?

Claude, Gemini and OpenAI each predict all 72 group-stage matches. As the real results come in, every correct call scores points. The Brier score measures how well each model's probabilities are calibrated — lower is better.

Group stage progress72/72 matches
WebBaselineEnriched

API, no tools + standardized context block (same data for every model).

Claude

leading
130
points · 72 predictions submitted
Exact scores
10
Correct results
42
Points accuracy
36.1%
Result accuracy
58.3%
Brier (calibration ↓)
0.516
Scored matches
72

Gemini

118
points · 72 predictions submitted
Exact scores
5
Correct results
42
Points accuracy
32.8%
Result accuracy
58.3%
Brier (calibration ↓)
0.525
Scored matches
72

OpenAI

123
points · 72 predictions submitted
Exact scores
8
Correct results
43
Points accuracy
34.2%
Result accuracy
59.7%
Brier (calibration ↓)
0.518
Scored matches
72

How scoring works

  • +5Exact score
  • +3Correct result + one exact side
  • +2Correct result only (W/D/L)
  • +0Wrong result

Experiment design

See the prompts →
  • Web— Chat + live web access — free sourcing.
  • Baseline— API, no tools — internal model knowledge only.
  • Enriched— API, no tools + standardized context block (same data for every model).