Madison Keys, Qinwen Zheng and a Model Miss
Qinwen Zheng overturned Madison Keys' favored status with a comeback that tested a shared pre-match view.
The favorite lost control
The model was wrong about this match. Madison Keys began as the stronger pre-match pick, but Qinwen Zheng turned a one sided opening into a comeback that changed the tenor of the whole contest.
Keys took the first set 6-1 and appeared to have imposed the clean, direct tennis that made her the favorite. Zheng then claimed the second set 6-7(3), keeping the match alive after a tiebreak that denied Keys a straight-sets finish. That was the hinge. A match that had looked settled became a deciding set with pressure moving to the player expected to close.
Zheng's recovery went further still. She came back from 5-0 down in the decider and survived match point before taking the match 5-7. Few results expose the limits of a pre-match estimate more sharply than that sequence, because the estimate described the starting position rather than a promise about every swing within the match.
The final score was 6-1 6-7(3) 5-7. It captures two very different versions of the contest: Keys' early authority, then Zheng's refusal to let the deficit become the ending. The favorite had the better forecast, but Zheng produced the points that mattered after the match had become most unstable.
What the model expected, and what happened
Article snapshot, Sep 6, 2026 · Not live
Qinwen Zheng won 1-6 7-6(3) 7-5. The model had made the other player the favorite.
| Player | Model before the match | Result |
|---|---|---|
| Madison Keys | 59% | Lost |
| Qinwen Zheng | 41% | Won |
Why the pre-match case favored Keys
Keys had a solid case before play began. She arrived ranked No. 17, with a career best of 7, while Zheng was ranked No. 23. Rankings do not decide a single afternoon, yet they helped explain why the match began with Keys carrying the larger expectation.
Their recent records did not separate them. Each player had won 6 of her last 10 matches, and each had posted the same 6-4 record across her latest hard-court sample. That made this less a form mismatch than a close meeting in which Keys' broader standing gave her the edge.
There was little rest disparity either. Both players had last competed 2 days earlier and were described as fresh, removing an obvious scheduling explanation for either player. Zheng had played 4 matches over the previous fortnight, while Keys had played 2, but neither entered with a long absence hanging over the matchup.
Keys also entered on a two-match winning run. Zheng's run was longer at four matches, which gave her a current rhythm argument that the final result ultimately strengthened. Still, neither streak was enough to overturn the larger pre-match view, especially with the hard-court records level.
A rivalry with no clear owner
This was never a matchup with a settled pattern. The head-to-head stood at 1-1 before the US Open, and it was also 1-1 on hard courts. Those records made a Keys win plausible, but they offered no basis for treating Zheng as a distant outsider.
Their most recent meeting had already shown Zheng's ability to reverse momentum against Keys. In Miami, Zheng won 4-6 6-2 6-4 after dropping the opening set. That earlier result did not determine what would happen in New York, though it did make the shape of this comeback less surprising in retrospect.
The US Open match echoed that Miami reversal in a more dramatic form. Keys again established the early advantage, then Zheng found a route back as the contest lengthened. The scoreline makes the contrast plain, because Zheng lost the first set heavily before winning both sets that followed.
That is why the outcome should not be read as evidence that the matchup was obvious all along. An even head-to-head can contain very different match scripts, and this one included a near finish before Zheng's escape. Keys had enough early control to support the forecast, while Zheng had enough resilience to defeat it.
What the model says
The model gave Keys a 59% chance and Zheng a 41% chance before the match. That made Keys the selection, but it also left a meaningful path for a Zheng win in a best-of-three contest.
The market consensus also had Keys at 59% and Zheng at 41%, so there was no gap between the model and the market at the start. Earlier sentiment had placed Keys at 63%, then moved 4 points toward Zheng before play began. The move anticipated a tighter contest without making Zheng the favorite.
That agreement matters because this was not a case of the model standing alone against the prevailing view. Both assessments saw Keys as more likely to win, and both were beaten by a match that turned on Zheng's late recovery. The result is a reminder that a probability is a range of possible outcomes, not a verdict.
The next round would bring Zheng in after an emotionally demanding comeback, with her four-match winning run now carrying extra weight. Keys leaves New York after holding a match point and a commanding lead, which makes the loss especially difficult to separate from the quality of her opening tennis.
Explore this matchup
Compare both players side by side: recent results, head-to-head and the model.
Player comparison
Article snapshot, Sep 6, 2026 · Not live
Each measure on the same scale: recent wins, wins on this surface, head-to-head wins and the model's chance.
| Measure | Madison Keys | Qinwen Zheng |
|---|---|---|
| Wins, last 10 matches | 60% (6-4) | 60% (6-4) |
| Wins, last 10 on hard | 60% (6-4) | 60% (6-4) |
| Head-to-head wins | 50% (1-1) | 50% (1-1) |
| Model win probability | 59% | 41% |
Head-to-head by surface
Article snapshot, Sep 6, 2026 · Not live
Madison Keys and Qinwen Zheng have met 2 times. The record is level at 1-1.
| Surface | Madison Keys | Qinwen Zheng |
|---|---|---|
| All surfaces | 1 | 1 |
| Hard | 1 | 1 |
Model and market on Madison Keys
Article snapshot, Sep 6, 2026 · Not live
Madison Keys's chance as the market consensus priced it from the opening line, next to the Court Shark model.
| Reading | Madison Keys | Qinwen Zheng |
|---|---|---|
| Opening consensus | 63% | 37% |
| Consensus at the start | 59% | 41% |
| Court Shark model | 59% | 41% |
Court Shark win probabilities are statistical estimates, not advice. No outcome is guaranteed.
Methodology and sources
How we build model estimates and where our information comes from.
Model estimates and observed results
The win probabilities come from the Court Shark match model, trained on completed ATP and WTA matches. A probability says how often a player in this position wins, not that one result will happen. The model's number for a match is recorded before play and never changed afterwards, so it can be checked against the result.
Sources and citations
Results, scores, rankings and serve statistics come from the Court Shark match database. The market consensus is the pre-match price for each player converted to a probability. Figures in this article are a snapshot taken when it was written.
Limitations
Serve and return rates count only matches with full serve statistics, which are sparser for Challenger and lower-level events, and the tour average they are compared with covers every year on record. Head-to-head records count every meeting with a recorded winner, walkovers and retirements included. Charts and tables are a snapshot and do not update after publication.
About the desk
Who writes the Journal and how an article is put together. Read about the desk
Editorial standards
What we publish, what we check and how we decide what is fit to print. Read our standards
Corrections
Spotted an error? Write to hello@courtshark.ai and we will correct it and note the change. Read the corrections policy