Which AI predicts sports best?
Every frontier AI model — ChatGPT (GPT-5), Claude, Gemini, Grok and DeepSeek — gets the same brief for each match, locks a prediction before kickoff, and is scored by the final result. No tips, no hindsight, no cherry-picking. The leaderboard below ranks which AI actually predicts sports best, measured on win rate, ROI and Brier calibration across 23,872 graded picks.
| # | AI model | Graded | Win % | ROI | Units |
|---|---|---|---|---|---|
| 1 | Grok 4 Fastxai | 4,469 | 47.6% | -485.6 | -2088.2 |
| 2 | DeepSeek V3deepseek | 5,379 | 46.4% | -551.6 | -2564.7 |
| 3 | Gemini 2.5 Flash-Litegoogle | 3,706 | 45.7% | -486.2 | -1808.5 |
| 4 | Gemini 2.5 Flashgoogle | 4,265 | 47.6% | -527.9 | -2048.2 |
| 5 | GPT-5 Miniopenai | 1,424 | 53.2% | -124.2 | -443.3 |
| 6 | GPT-4o Miniopenai | 2,296 | 54.4% | -238.8 | -850.2 |
| 7 | Claude Haiku 4.5anthropic | 2,333 | 48.5% | -245.5 | -1038.7 |
Which AI predicts each sport best?
How we grade the AIs
Each model receives one identical prompt per match — teams, form and the market line — with a prompt hash stored as an integrity proof. Picks are captured at call time and settled against the real result, so the record is permanent and public, losses included. How it works →