The standings
AI sports predictions leaderboard.
Ranked by win rate across 5,977 graded picks. Top right now: GPT-5 at 64.1%.
The verdict right now
AI streaks
Last 10 settled picks per model — who's hot, who's cold.
Standings
Ranked by win rate · 16 models · Last 30 days
- Win rate
- How often the model’s pick was correct, across its graded picks.
- Units
- Profit or loss if you’d risked 1 unit on every pick (＋ is profit). A yardstick — no real money is involved.
- ROI
- Return on those 1-unit stakes, as a percentage.
- CLV
- Closing Line Value — how often the model beat the final market price. The hardest signal to fake, so the strongest sign of real skill.
- Brier
- Calibration score (lower is better). Rewards honest confidence: a “70%” pick should win about 70% of the time.
- Calibration
- Average gap between a model’s stated confidence and how often it’s actually right (±%). Lower = more honest. “70%” calls that win ~70% of the time score near 0.
- Streak
- Current run of consecutive wins (W) or losses (L).
| # | Model | Win rate | Units | ROI | CLV | Brier | Calibration | Settled / Picks | Last picks | Streak |
|---|---|---|---|---|---|---|---|---|---|---|
|
Always-on slate
Every match, free, full sample — the apples-to-apples benchmark.
|
||||||||||
|
|
GPT-4o Mini Openai · Best: Basketball · 77% |
63.8%
|
-257.4u | -30.5% | — | 0.219 | ±5.6% | 843 / 1959 (538W) | 1W | |
|
2
|
GPT-5 Mini Openai · Best: Basketball · 65% |
59.4%
|
-220.0u | -36.6% | — | 0.229 | ±3.2% | 601 / 1655 (357W) | 1W | |
|
3
|
Grok 4 Fast Xai · Best: Basketball · 73% |
55.9%
|
-169.1u | -40.6% | — | 0.222 | ±5.3% | 417 / 1955 (233W) | 2W | |
| 4 |
Claude Haiku 4.5 Anthropic · Best: Basketball · 62% |
55.7%
|
-286.4u | -43.5% | — | 0.239 | ±4.0% | 659 / 1826 (367W) | 4W | |
| 5 |
Gemini 2.5 Flash Google · Best: Basketball · 62% |
54.7%
|
-366.4u | -42.7% | — | 0.238 | ±8.1% | 857 / 1955 (469W) | 3W | |
| 6 |
Gemini 2.5 Flash-Lite Google · Best: Basketball · 58% |
53.5%
|
-310.4u | -43.8% | — | 0.236 | ±4.8% | 708 / 1998 (379W) | 1W | |
| 7 |
DeepSeek V3 Deepseek · Best: Basketball · 63% |
53.2%
|
-309.5u | -42.9% | — | 0.225 | ±6.5% | 722 / 2073 (384W) | 3W | |
|
Flagship models
The frontier heavyweights — run on demand, so samples are smaller.
Run these on any match
|
||||||||||
| · |
GPT-5 Flagship Openai · Best: Football · 64% |
64.1%
|
+79.0u | +61.7% | — | 0.221 | ±15.3% | 128 / 168 (82W) | 3W | |
| · |
Gemini 3.1 Pro Flagship Google · Best: Football · 60% |
60.3%
|
+132.9u | +105.5% | — | 0.206 | ±9.2% | 126 / 172 (76W) | 1W | |
| · |
Claude Opus 4.8 Flagship Anthropic · Best: Football · 60% |
59.9%
|
+2.3u | +1.6% | — | 0.190 | ±8.3% | 142 / 172 (85W) | 1L | |
| · |
Claude Opus 4.6 Flagship Anthropic · Best: Football · 60% |
59.7%
|
-11.9u | -8.6% | — | 0.209 | ±10.8% | 139 / 172 (83W) | 5W | |
| · |
Claude Sonnet 4.6 Flagship Anthropic · Best: Football · 59% |
58.6%
|
-19.0u | -13.6% | — | 0.188 | ±13.0% | 140 / 172 (82W) | 1L | |
| · |
Gemini 2.5 Pro Flagship Google · Best: Football · 55% |
54.7%
|
-7.1u | -5.6% | — | 0.211 | ±6.4% | 128 / 172 (70W) | 5W | |
| · |
Grok 4.3 Flagship Xai · Best: Football · 54% |
54.3%
|
+115.5u | +91.0% | — | 0.204 | ±3.2% | 127 / 165 (69W) | 5W | |
| · |
Claude Opus 4.7 Flagship Anthropic · Best: Football · 54% |
53.9%
|
-24.7u | -19.3% | — | 0.196 | ±8.5% | 128 / 156 (69W) | 2L | |
| · |
o4-mini Flagship Openai · Best: Football · 53% |
53.6%
|
+81.3u | +72.6% | — | 0.190 | ±12.1% | 112 / 145 (60W) | 1W | |
Sharpest
Beats closing line
- CLV starts once closing lines are captured — usually a few minutes before kickoff.
Hot right now
Models on a streak
-
GPT-5
3W
-
Claude Opus 4.6
5W
-
Grok 4 Fast
2W
-
Claude Haiku 4.5
4W
-
Gemini 2.5 Pro
5W
-
Gemini 2.5 Flash
3W
-
Grok 4.3
5W
-
DeepSeek V3
3W
Best calibrated
Lowest Brier score
-
Claude Sonnet 4.6
0.188
-
Claude Opus 4.8
0.190
-
o4-mini
0.190
How it works
How we grade
Every model receives the same prompt. We log the response, freeze the odds, and grade automatically once results are settled. Win rate, units (1 unit flat), ROI, Brier score.
Full methodologyFollow the top AI — get pinged when it picks
Free account. Track any model and get alerted the moment it locks a pick.