How accurate is GPT-5.6 at football predictions?
openai
Real predictions, graded against real results — a live experiment, not a proven edge. Last 30 days, recomputed after every graded match (updated 22 Jul 2026, 03:27 UTC).
Result accuracy
65.1%
Exact score
18.1%
Calibration
53.3%
Graded predictions
83
01 — Confidence & calibration
Across 83 predictions with a stated confidence, GPT-5.6 averaged 48.6% confidence — 51.0% when it was right vs 44.3% when it was wrong. Calibration score: 53.3%.
02 — By competition (30 days)
| Competition | Graded | Result acc. | Exact score |
|---|---|---|---|
| World CupWorld | 63 | 66.7% | 20.6% |
| Liga MXMexico | low sample9 | 44.4% | 11.1% |
| Serie ABrazil | low sample6 | 66.7% | 16.7% |
| Major League SoccerUSA | low sample5 | 80.0% | 0.0% |
Low-sample rows are dimmed, not hidden — transparency beats tidiness.
03 — Betting experiment (30 days)
1 virtual unit per pick, tracked in units — a live experiment, not betting advice. 18+.
Win rate
59.0%
ROI
+15.1%
P/L
+12.6u
Avg odds
2.16
Bets
83
← Full model leaderboard · how grading works · compare prediction tools