A Jiri Lehecka vs Daniel Adolfo Vallejo prediction for Sunday's Tokyo quarterfinal starts with a serve clinic and ends with a number that should worry anyone backing the favorite outright. First serve at 4:30 AM ET, the model's side is Lehecka, and the market has no moneyline posted on this one as of our write time.
The tier on this one is a PASS. That's not a knock on the read, it just means this isn't a bet we're making today, and we'll leave it there rather than guess at which specific trigger kept us off it.
Why the model leans Lehecka
Lehecka's 12-month hard-court hold rate sits at 85%, and he's been running well above that line all week in Tokyo: 31 of 32 service games held, or 97%. That's not a fluke week either. He arrived fresh off beating Ben Shelton 6-4 6-4 and Learner Tien 6-3 7-5 in Davis Cup play, both wins coming days after Shelton's own US Open final, and both matches that count as earned rather than gift-wrapped. In Tokyo itself he beat Zizou Bergs in three and then Ugo Humbert in three, winning 80 to 91 percent of his first-serve points along the way. His softer three-month serve line, a 71.2% first-serve-won and a 49.8% second-serve-won number that look mediocre in isolation, has simply not shown up this week.
The case for Vallejo
Vallejo's Tokyo run is just as real. He's held 40 of 44 service games (91%), well clear of his thin 12-month hard-court mark of a 71% hold rate across a 6-6 record. The headline result is a 7-6 6-1 win over the bigger-serving Matteo Berrettini in the round of 16, where he saved all three break points he faced. His return game is the actual threat to Lehecka specifically: Lehecka's second serve has won only 49.8% of points over the past three months, which is close to a coin flip on its own, and Vallejo's return has been built all tournament around pressuring second serves. The catch is that his Tokyo surge followed a brutal stretch, a Davis Cup tie in Lima that included a clay loss to Juan Pablo Varillas, then Hangzhou, then qualifying rounds here. That clay result is a workload data point, not a form read on this hard court, and his run through Tokyo has come against a field that was, at different points, travel-worn or overmatched.
The break-point number that matters
Here's the piece that keeps this from being a clean call either way. Lehecka has broken serve in Tokyo just 4 of 30 times, a 13% break rate, and has converted only 4 of his 22 break-point chances, about 18%. He's winning his own service games at a near-perfect clip but is not doing damage on return, which means long stretches of this match are likely to come down to tiebreaks rather than clean service breaks. Vallejo, by contrast, has broken 14 of 42 chances in Tokyo, a 33% rate, which is the strongest part of his week. That's a notably stronger break number than Lehecka's, and he's produced it despite a much heavier workload: six singles matches and about 12 hours of tennis in two weeks, against Lehecka's two matches and under six hours. With that much daylight between their break production and their rest, a tiebreak-heavy match can tip either way.
The verdict
The model's lean is Lehecka, built on a dominant week of first-strike serving and a recent form run that erases the inactivity flags in his longer-term numbers. But Lehecka's inability to convert break points all tournament, just 4 of 22, means he isn't creating the separation his serve numbers would suggest, and Vallejo's own hot hold rate and sharper return against second serves keep this live deep into sets. That gap is exactly why this is a PASS. We're not backing either side here, and we're not naming the specific threshold that kept us off it.
- Event · Tokyo · ATP (2026) · Sunday, October 4, 2026 · 4:30 AM ET
- Model · T Money, Botcappers' AI Tennis model — how it is built and what it does not know
- Record · every pick is timestamped before the event and graded in public, win or lose
- Statistics · the model's own match/game database, current to the posting time above. Figures a reader cannot open here are internal to that database, not published third-party data.

