How Accurate Are Tennis Predictions? Our Real Numbers

In 2026 our published ATP and WTA moneyline picks have gone 325 for 385, a hit rate of 84.4%. That is not a demo number. It is the running score of matches we called and posted before the result was known. The year before, in 2025, the same picks landed 649 of 797, or 81.4%. Two seasons, one steady number in the eighties. Add the game-spread calls on the same tours, which settle on a line rather than on a winner, and 2026 reads 509 for 711, or 71.6%, which is the blended figure the predictions page shows.

Scope: unless marked as blended, every figure in this article covers ATP and WTA moneyline picks that clear the public track-record filter behind our predictions page, as of 2026-08-26. Game-spread picks settle on a line rather than on the winner, so they are counted separately, and ITF results sit on their own tabs. The headline on the predictions page blends moneyline and game-spread picks, which is why it reads lower than the moneyline-only rates here, and it moves as new results settle.

Accuracy is an easy word to abuse in this sport, so it is worth being exact about what we are counting. We score a pick as correct when the player we favored wins the match. Nothing else. Not sets, not games, not whether the price moved our way. The winner of the match either was the player we named or was not, and the tally follows from there. When we quote a hit rate, that is the only thing behind it.

What the number is not

A high pick-the-winner rate is not the same as a betting edge, and we try never to blur the two. Odds here always mean the book's price. A model can call the favorite correctly four times out of five and still lose money if it never disagrees with the market. So read our accuracy as a measure of how often we point at the right player, not as a promise about returns. The two questions are related but they are not the same question.

The honest bar for any tennis model is simple: back the bookmaker's favorite in every match and see how often you are right. Most matches have a clear favorite, so that naive rule already gets a lot correct on its own. A model earns its keep only by clearing that bar and then finding the spots where the obvious call is wrong. Beating the favorite baseline is the whole job.

Recent form

The long-run seasons tell you the model is stable. The short window tells you it is still working right now. That is why the predictions page carries a rolling recent window next to the season rows rather than a number frozen into an article. Short windows are small samples and they drift back toward the season figure, the way small samples always do. We publish the window anyway, because a model that quietly stops matching its own history should show up there first.

ATP and WTA are not the same problem

Split the record by tour and the two circuits behave a little differently. These next rows cover the full published moneyline record, back to 2021. Every pick in them was generated walk-forward, meaning the model only ever saw matches that happened before the one it was scoring, so it never learned from the future.

The WTA record lands higher, at 83.7%, but on a far smaller sample of 381 matches, so read that gap with some caution rather than as proof one tour is easier. The ATP figure of 80.9% rests on 3860 matches and is the more settled of the two. Both sit in the same neighborhood as the current season, which is the reassuring part: the model performs on new matches about as well as its history says it should.

Where the accuracy number misleads

An 84.4% hit rate sounds close to solved, and it is not. Most of that number is the model agreeing with the obvious favorite, which anyone can do. The matches that decide whether a model is genuinely good are the coin-flips and the upsets, and no single accuracy figure tells you how it does there. A model could post a strong headline rate and still be useless on exactly the matches where you wanted help.

Accuracy also says nothing about price. Being right 84.4% of the time means little if the market already had the favorite at a price that leaves no room. That is why we keep the accuracy story and the betting story in separate boxes, and why the pick-the-winner rate is the start of the analysis, not the end of it.

You can watch the live blended track record, including game-spread picks, on today's predictions, and the ratings that feed those calls sit next to the official points on the ATP rankings page.

The honest read: our moneyline picks have hit the eighties for two straight seasons, but the headline rate is mostly the easy matches. Judge a model by what it does with the hard ones.