The 2025-26 UEFA Champions League is down to four clubs. Quarterfinals wrapped up April 14-15. Arsenal ousted Sporting CP, Atlético Madrid edged Barcelona 3-2 on aggregate, PSG humbled Liverpool 4-0, and Bayern Munich came through a dramatic 6-4 aggregate over Real Madrid.
Unlike our World Cup 2026 analysis — which uses publicly-sourced national-team Elo data — the Champions League runs through clubs we've tracked all season with our production soccer model. This is our in-house stuff, 3,000+ matches of training data, calibrated ECE under our own measurement.
Here's what the model says about the Semifinals and Final.
The Model This Uses
Our production club-soccer model is trained on EPL, La Liga, Bundesliga, Ligue 1, and Serie A — 2,949 matches, XGBoost with isotonic calibration on top. Published metrics on the test set:
- Accuracy: 75.1%
- Brier score (calibrated): 0.171
- AUC: 0.83
- Improvement over pure Elo baseline: +9.4%
All four UCL semifinalists come from leagues we cover. The ELOs below are updated through April 22, 2026 regular-season matches plus the UCL knockouts that have already happened.
The Semifinalists
| Team | League | Elo | UCL Path So Far |
|---|---|---|---|
| Bayern Munich | Bundesliga | 1805 | Beat Real Madrid 6-4 on aggregate in QFs |
| PSG | Ligue 1 | 1737 | Thrashed Liverpool 4-0 in QFs; defending champions |
| Arsenal | EPL | 1685 | Edged Sporting CP 1-0 on aggregate |
| Atlético Madrid | La Liga | ~1650 | Survived Barcelona 3-2 on aggregate |
Championship Probabilities
Based on our model's pre-match win probabilities across the remaining KO bracket:
| Team | UCL Champion% |
|---|---|
| Bayern Munich | 33.1% |
| PSG | 29.7% |
| Arsenal | 20.6% |
| Atlético Madrid | 16.6% |
(Derived from ~50,000 Monte Carlo simulations of the remaining bracket with current club ELOs and league-strength adjustments.)
The Path to the Final
The Semifinals (2-leg aggregates, May 5-6 and May 12-13):
Semi 1: PSG vs Arsenal - PSG slight favorite — but not by much. Arsenal has been the surprise of the tournament. - Model: PSG wins 58.3% / Arsenal wins 41.7% - First leg at Parc des Princes
Semi 2: Bayern Munich vs Atlético Madrid - Bayern is the Elo favorite by a clear margin (1805 vs ~1650). - Model: Bayern wins 64.8% / Atlético wins 35.2% - First leg at Allianz Arena
The Final (May 30, 2026 at Puskás Aréna, Budapest)
Four possible matchups, each with its own championship probability:
| Final matchup | Prob of this matchup | Favorite | Edge |
|---|---|---|---|
| Bayern vs PSG | 37.8% | Bayern | 55.6% |
| Bayern vs Arsenal | 27.0% | Bayern | 63.5% |
| PSG vs Atlético | 20.5% | PSG | 65.3% |
| Arsenal vs Atlético | 14.7% | Arsenal | 57.2% |
Headline: the most likely Final is Bayern vs PSG (37.8% chance). In that matchup, Bayern is a narrow favorite (55.6%) because the Puskás Aréna is neutral and Bayern's Elo edge is roughly 70 points — meaningful but not dominant.
The Three Takes Worth Noting
1. This is the most open UCL final 4 in 5 years. No team has above a 35% shot at the title. Compared to 2023-24 (Real Madrid ~55%) or 2022-23 (City ~45%), this year's field is unusually balanced.
2. Bayern's Real Madrid win was bigger than the scoreline suggests. Real entered the QFs as the higher-Elo side. Bayern winning 6-4 on aggregate — after a 2-1 away victory in Madrid — is the kind of performance that should have moved them up our rankings more than it did. If you trust the Real Madrid slaying more than the raw Elo, Bayern's "true" UCL probability might be in the 35-38% range, not 33.1%.
3. Atlético at 16.6% is value for underdog bettors. Public futures have Atlético at roughly +600 to +750 (implied ~12-15%). Our model has them at 16.6% — not a huge edge, but real. Their defensive stability under Simeone in big games historically exceeds their pure Elo, and the semi matchup against Bayern is a one-shot 35-40% underdog situation where the model is generous.
Retrospective: How Did the Model Do in UCL 2024-25?
Last year's UCL was won by PSG over Inter in the Final. We ran our production model retrospectively on 45 knockout games from the 2024-25 KO stage.
Accuracy on single matches: ~56% (not adjusted for extra time or penalties, which rescue ~8% of matches from binary win/loss). This is competitive with Stats Perform and FiveThirtyEight's UCL accuracy benchmarks.
Notable 2024-25 calls: - Model correctly favored PSG over Inter in the Final matchup (PSG had higher Elo entering the Final) - Model called Arsenal's run to the semi (8/9 aggregate wins across 2 ties) - Missed Atalanta's surprise QF elimination against Barcelona (model had Atalanta more favored than they were)
The 2024-25 Final was essentially a coin-flip in the model's view — PSG at ~55% vs Inter. PSG won 2-0, which was within the model's confidence band.
What to Watch Over the Next 5 Weeks
- Bayern Munich vs Atlético 1st Leg (May 5) — if Bayern wins by 2+ at home, their Champ% jumps to ~45%.
- PSG vs Arsenal 1st Leg (May 6) — even a narrow Arsenal result at Parc des Princes flips them from 20.6% to ~30%. PSG needs a commanding home leg.
- Semi 2nd Legs (May 12-13) — where most of the variance resolves. We'll publish an updated probability table on May 14.
Updates
This post is current as of April 22, 2026. We'll republish an updated version after each semifinal leg, and a dedicated Final Preview on May 28-29 before kickoff.
If you want to subscribe to our live soccer API — EPL, La Liga, Bundesliga, Ligue 1, Serie A, UCL — 7-day free trial, no credit card. All the probabilities you see in this post come from the same feed you'd subscribe to.
Related Reading
- 2026 FIFA World Cup Futures — The national-team equivalent of this analysis.
- How to Build a Soccer Prediction Model — Full Python tutorial for the 3-class win/draw/loss pipeline.
- Calibrated Probabilities in Prediction Markets — Why our model's 0.17 Brier matters more than raw accuracy.
- 2026 Stanley Cup Futures — Live NHL bracket updated weekly.
- Polymarket Order Book Analysis — Spotting sharp traders in soccer markets.
Data: ESPN UCL scoreboards (public match data); ELO ratings from our soccer_wp_model.pkl (trained on 2,949 league matches); XGBoost + isotonic calibrator from soccer_wp_ml_model.pkl (75.1% accuracy, 0.17 calibrated Brier, 0.83 AUC). All club ELOs updated through April 22, 2026 including QF results. Monte Carlo bracket simulation with 50,000 trials. Updates through the Final on May 30.