We Priced the US Open Before the Draw Exists — Here's the Whole Board, Including the Bets We're Against

The 2026 US Open men's and women's brackets won't be drawn until a few days before play starts on August 30. Sportsbooks have had championship odds up for weeks anyway — and so have we. Predictium's tennis model has published a full 128-runner outright board for both singles draws since late July, with our probability, the market's price, and the expected value of every runner the books list. Not just the ones we like.

A tennis major is a 128-player knockout. Once the draw exists, our model prices it exactly: serve-and-return skill estimates feed a point-by-point match engine, and a bracket calculation rolls every possible path to the title into one championship probability per player.

Before the draw exists, there's a wrinkle: nobody knows who plays whom. Our answer is to simulate the draw ceremony itself. Grand Slam seeding rules pin the top seeds to opposite halves and randomize everyone else within constraints, so we sample thousands of legal brackets, price each one exactly, and average. The result is each player's championship probability over every draw they could plausibly get.

Does the missing bracket matter? Less than you'd think. Fields at this depth are deep enough that a kind draw and a cruel draw mostly average out, and in our own testing, knowing the real bracket moves our pre-tournament read on the eventual champion by only about a tenth of a percentage point. When the real bracket lands on draw day, our numbers should barely move — and we've frozen the pre-draw board so you can check that claim against what we published weeks early, rather than take our word for it.

Add up the implied probabilities on a sportsbook's US Open outright board and you don't get 100%. On the boards we're pricing against right now, you get anywhere from 101% to 174%, depending on the venue and how many runners it lists:

That margin has to be stripped out ("de-vigged") before a posted price says anything about a player's real chances, and here's the uncomfortable part: there is no single correct way to strip it.

Take Jannik Sinner on FanDuel's men's board. His raw price implies a 61.5% chance of winning the title. De-vig that board three standard ways and you get three very different answers: 37.5% (proportional), 54.4% (power), or 60.9% (additive). That's a 23-point spread on the same price, from nothing but a choice of arithmetic. The math you never see is doing more work than the odds you do.

That's why our board shows two EV columns instead of blending everything into one number:

• EV Kalshi — computed against the price on Kalshi, a CFTC-regulated exchange where people trade against each other at roughly 1% margin. Almost nothing to strip out, so its quotes sit closest to a true fair price. The catch: it only lists a dozen or so contenders per draw.
• EV Book — computed against the best sportsbook price available. Full 128-runner coverage, but every quote is padded by a much fatter margin.

Each column is the same arithmetic — our probability multiplied by that venue's actual price, minus one — so when the two disagree it's because the venues are quoting genuinely different numbers for the same outcome, not because we changed our mind between columns. Showing both is how you see that, instead of us quietly picking one.

As of August 8, 03:18 UTC. This board refreshes up to three times a day, so treat these as a snapshot and the live boards as the record.

Men's

Women's

Each EV is our probability times that venue's own price, minus one. A dash means the player isn't listed on the exchange. † Alcaraz is a special case — see the next section.

Where you bet changes the answer. Take Zverev: we make him a 7.8% champion, the sportsbooks offer 9.00 and the exchange offers 10.00 for the identical outcome. Our opinion doesn't move between those two cells — but the EV improves from −30% to −22% purely because the exchange isn't padding a margin into its price. Neither is a bet. It's just a reminder that a single blended EV number on a futures market tells you less than it pretends to.

Plenty of the board is negative, including the top of it. Sinner is the men's favourite by a mile, and we agree he's the most likely champion — but we make him 45.2% where the market's de-vigged price implies about 58%, so at today's number he's a losing bet by our math and the board says so, in red, at the top. Djokovic grades around −62%, Andreeva around −53%, Osaka around −68%. A futures page that only shows green numbers is a tout sheet. Ours shows the reds, because the reds are what make the greens mean something.

The women's board is where we disagree most with the market. We make Rybakina a 16.1% champion where the exchange's de-vigged price implies about 7.9% — more than double. Swiatek and Pegula show the same shape. We're not softening those: one of us is wrong, the numbers are published, and they'll be graded. The same goes for de Minaur, whom we rate a top-five men's contender while the books price him around twentieth.

Look at Alcaraz's line again. Our model makes him a 19.2% champion. The exchange prices him near 12%. One sportsbook has dropped him from its outright board entirely, and the one that still lists him has drifted him out to +310.

Naively, that reads as a large positive edge against the exchange. It isn't, and this is the most instructive row on either board.

Alcaraz withdrew from the Cincinnati Open on August 4 with an ongoing wrist injury. He hasn't played a competitive match since Barcelona in April, having already missed Roland-Garros and Wimbledon, and his defence of the US Open title is genuinely in doubt — he may arrive with no match play since spring.

The market has priced that. We haven't. Our model estimates serve and return skill from matches that actually happened, and Alcaraz's most recent one is about four months old. It knows exactly how well he plays tennis and nothing whatsoever about his wrist. So it prices a healthy Alcaraz, and the exchange — where people with opinions about that wrist are trading against each other at a one-cent spread — prices the real one.

That gap is not an edge. It's unpriced participation risk, and it's ours, not the market's. Quoting a number against people who can see something our inputs can't is the one situation where our own EV column is the least trustworthy thing on the page.

The right response isn't to hide the row or to quietly widen a threshold until the number goes away. It's to say plainly that our data is too old to have an opinion here — which is precisely the discipline our match-level board already applies when a player's information goes stale, and which we're extending to futures. When that lands, Alcaraz's EV cells will read "no claim," with a caption explaining that the silence is about our data, not his chances. His probability and the market's price will keep publishing side by side, because the disagreement is the useful part.

We checked the rest of both boards for the same exposure. Alcaraz is the only claim-carrying runner on either draw who hasn't played in over a month. One row — but worth the whole section, because a model that can't tell you when to distrust it isn't worth much.

Every runner we give under a 1% chance of winning the title shows "no claim" instead of an EV — 95 of them on each board. At 200-to-1, a tiny error in our probability becomes a giant swing in EV; the percentage isn't information down there, it's amplified noise.

The threshold is on our own probability, not the market's price, so a player doesn't lose their claim just because a particular venue declined to list them, and a longshot the market loves doesn't get a number from us just because it's cheap.

What we deliberately do not do is suppress a number for merely disagreeing with the market. On outright markets, big model-versus-market gaps at long prices are normal, and hiding our widest disagreements would flatter our track record by deleting exactly the claims most worth grading. Rybakina at more than double the market's implied number stays on the board. Alcaraz comes off it for a different reason entirely: not that we disagree, but that we know our input is stale.

Everything on the futures board follows the same rule as the rest of Predictium: walk-forward only — every published probability comes from a model trained strictly on data before the events it's predicting, and every methodology claim is backtested the same way.

The full futures backtest is public, season by season, on our futures backtest page: 891 tournaments, 2018 through 2025. Across all of them, our average pre-tournament probability on the eventual champion is 6.45%, against 5.98% for an Elo-only bracket baseline and 2.64% for a field-blind one. The model beats the Elo baseline in every single season, 2018 through 2025 — not on average, every year. On Grand Slams specifically, where the fields are deepest and the seeding most protective, we put 10.6% on the eventual winner on average, pre-tournament.

And the pre-draw prices you're reading today are frozen the day the real bracket is committed. When the draw lands, you'll be able to compare what we said weeks early against what we say with the bracket in hand.

Nothing here is betting advice. It's a model, its disagreements with the market, and a public record of both — including the row where the market is probably right and we're probably stale.

Boards update several times daily through the start of the tournament: Men's board · Women's board · How the tennis model works