The Tennis Model Is Live

Our tennis match-winner model runs on a fully rebuilt pipeline as of today. It covers every ATP and WTA main-draw match on the slate, retrains on fresh results every morning, and prices each match with a calibrated win probability.

The methodology is proprietary, but the core idea is player ratings. Every completed match on both tours updates each player's rating, with surface taken into account, and those ratings feed a model that outputs a win probability for every scheduled match. One thing worth knowing: the model never looks at betting odds. Its probabilities are built entirely from match results, so when it disagrees with the market, that disagreement is genuinely independent information.

Trust the moneylines. That's it, for now.

Same message as our WNBA release, and we will keep repeating it: the match-winner picks are the product. Season to date the model is 185–89 picking winners, a 67.5% hit rate across every graded match on both tours. High-confidence picks, where the model gives its side at least a 62% chance, are 138–43, a 76.2% hit rate. Every pick is graded against final results and the full history stays on the page, wins and losses both.

Everything else is still in development. Set betting, game totals, and player props are not on the board because they have not met our bar yet. The simulation work behind them is in progress, and when they are ready you will read about it here first. Until then, the moneyline column is the one to use.

What's on the board

• Win probability and pick: the model's probability for each side, with the pick being whichever player comes out above 50%. This is the column to use.
• Confidence tier: every pick lands in a tier based on how strong the model's conviction is. Higher tiers have hit at a higher rate historically, and the tier records are graded separately.
• Calibration bands: each probability comes with an empirical error band built from thousands of past predictions in the same range, so you can see how often a "70% favorite" has actually won.
• Daily retrain and regrade: the model retrains on the latest completed matches every morning, and a rolling window of recent picks is regraded as official results come in.