Sports Betting Model Performance Statistics 2026
What does a quantitative sports betting model actually return? Below is our own answer, from an append-only public ledger — every pick recorded at publication, resolved automatically against official scores, losses included — as of 2026-10-16.
Quick answer: Across 2,955 resolved picks since November 2025, the Olympus Bets Analytics Monte Carlo models have won 51.9% (1,494W–1,385L) for -23.15u — roughly breakeven at the blended level, with the edge heavily concentrated by league: the best league is more than 80 units ahead of the worst. For context, a bettor needs ~52.4% at standard -110 pricing just to break even. The current premium regime (Core Premium, last 90 days) runs 51.4% and -8.87u across 258 picks. Every number regenerates daily from the same ledger.
Key Statistics (Citable)
- A quantitative betting model resolved 51.9% winners over 2,955 published picks — near the -110 breakeven threshold of 52.4% — per the Olympus Bets Analytics public ledger.
- League-level results diverged by more than 80 units between the best and worst league on the same model infrastructure, per the same ledger — league selection mattered more than headline win rate.
- The free, publicly published tier resolved 51.9% across 2,025 picks for +6.01u (+0.3% ROI), per Olympus Bets Analytics.
- Model validation spans 148,855 resolved games across 12 leagues — far larger than the pick ledger, because calibration learns from every game, not just games with picks.
Performance by League
Every league runs a dedicated simulation engine, and results differ because sports differ in market efficiency and modelability. This is the full distribution — if you bet one sport, judge the model on that row, not the blended headline. Rows sort by units won; leagues qualify with at least one resolved pick.
| League | Wins | Losses | Win Rate | Units P/L | Engine |
|---|---|---|---|---|---|
| GOLF | 308 | 235 | 56.7% | +50.83u | Conditional Sim V3.2 |
| MLB | 367 | 306 | 54.5% | +31.43u | Pitch Sim V5.2 PA-Hybrid |
| CBB | 228 | 195 | 53.9% | +13.29u | Savant Ultra v5.0.1 |
| LOL | 56 | 39 | 58.9% | +9.88u | Championship v2.1 |
| NFL | 22 | 13 | 62.9% | +7.74u | NFL v2.0 |
| ESPORTS | 13 | 6 | 68.4% | +6.02u | ESPORTS |
| WBC | 0 | 1 | 0.0% | -0.50u | Pitch Sim (WBC bridge) |
| FIBAW | 3 | 6 | 33.3% | -1.09u | FIBAW |
| UFC | 0 | 2 | 0.0% | -3.00u | Oracle PM signals |
| WNBA | 25 | 28 | 47.2% | -3.77u | Possession MC Four Factors |
| WC | 6 | 9 | 40.0% | -4.56u | V16.3 PBP xG |
| OLYMPIC | 23 | 30 | 43.4% | -10.95u | Olympic V2.2 |
| NHL | 90 | 82 | 52.3% | -15.40u | NHL v19.5 |
| TENNIS | 91 | 96 | 48.7% | -18.50u | Tennis V2 (calibrating) |
| CS2 | 38 | 83 | 31.4% | -19.55u | CS2 (collecting) |
| SOCCER | 66 | 89 | 42.6% | -29.70u | V16.3 PBP xG |
| NBA | 158 | 165 | 48.9% | -35.31u | Possession MC V5.0.2 |
Two honest caveats. First, samples differ enormously by league — a 55% win rate over 550 picks and a 60% win rate over 30 picks are not comparable evidence. Second, this table blends free and premium tiers and every confidence level; the track record page separates them, including rolling 30/60/90-day windows for the current regime.
What "Model Accuracy" Should Mean
Most published claims about betting model accuracy quote a win rate with no denominator, no odds context, and no losses. Three rules make the statistic meaningful:
- Win rate is meaningless without the breakeven rate. At -110, 52.4% breaks even. At heavy favorite prices, a 70% win rate can lose money. Units and ROI carry the information win rate alone cannot.
- The denominator must be every published pick. Our ledger is append-only: picks are recorded at publication with line, odds, stake, and model probability, then resolved automatically against ESPN scores. Nothing is added or removed after the fact.
- Validation should cover all games, not just picks. Calibration measured only on games the model chose to bet is survivorship-biased. Our validation store covers 148,855 resolved games — every simulated game, picked or not.
The methodology page covers the simulation, calibration, and sizing pipeline in detail; the AI sports betting statistics page quantifies the machinery itself.