Who will win the Premier League in 2026-27?
Manchester City and Arsenal are level on points and share 93% of the title. The model separates them on one detail — and England's best defence is not the one you would guess.
By Lucas da Silva · September 20, 2026 · 7 min read
Manchester City and Arsenal have 15 points each after five matchdays. They occupy the top two places, split by goal difference. On paper, nothing separates them.
The model separates them by twenty-six points of probability.
 Manchester City | 59.5% |
 Arsenal | 33.7% |
 Brighton & Hove Albion | 2.8% |
 Brentford | 1.2% |
Between them, City and Arsenal take 93% of the titles. Third-placed Brighton get 2.8%. The richest league in Europe is, this season, a two-horse race.
Where the gap between them comes from
Not from the defence. That is the counter-intuitive part of this projection.
| Arsenal | 0.68 |
| Manchester City | 0.74 |
Arsenal concede less than City: an index of 0.68 against 0.74. That is the best defence in England, and comfortably the better of the two.
The gap is at the other end: 1.51 for City against 1.37 for Arsenal. About ten per cent more attacking output, which translates into four projected points across the season — 80.3 against 76.4.
The final-position heat map
Where each team finishes
The probability of finishing in each position, across the 50,000 simulated seasons. A tight row is a team whose fate is nearly settled; a spread-out row, a team for whom anything is still possible.
Manchester City58279Arsenal3140167Brighton & Hove Albion714211511865
Liverpool717171411975Brentford10131212109765
Manchester United69101110109776
Leeds United6891010998765
Newcastle United79999988765
Chelsea578899988765
Everton567899998765
AFC Bournemouth567899109876
Aston Villa566788888765
Nottingham Forest567788888765
Hull City56778999987
Sunderland5689101111119
Fulham56810121313118
Crystal Palace6791113131310
Tottenham Hotspur569121418208
Ipswich Town57913182611
Coventry City61269The darker the cell, the likelier the position. Teams are ordered by average expected position, not by current standing.
Spectacular at the top and the bottom, unreadable in between.
At the top, two cells dominate: City first in 59% of runs, Arsenal second in 42%. Neither club practically ever appears below fourth.
At the bottom, Coventry City finish last in 65% of simulations. That is the darkest cell on any of our maps. A club whose attacking index is 0.47 — they score less than half an average side — is not saved by one good month.
In between, from fifth to fifteenth, no cell exceeds 12%. Eleven clubs for eleven places, and the model cannot say which goes where.
Liverpool and Manchester United, two table anomalies
England's two most decorated clubs are tenth and thirteenth. The model rates them fourth and sixth.
Where each team is heading
Average expected position at each matchday. Left of the marker, what actually happened; right of it, the projection.
Liverpool5.4eManchester City1.5eManchester United8.0eArsenal2.0eLiverpool project 58.7 points and a 26% chance of Champions League qualification. United follow on 23%.
Put another way: their start has been poor, their level has not. But the difference between "fourth-strongest" and "fourth place" has to be paid for, and five lost matchdays cost double to recover when seven clubs are chasing the same three tickets.
The top-four race, the only real battle
This is where the English season is decided.
| Club | Top 4 | Strength rank | Current position |
|---|---|---|---|
| Manchester City | 97.7% | 1 | 1 |
| Arsenal | 93.9% | 2 | 2 |
| Brighton | 43.5% | 3 | 3 |
| Brentford | 30.9% | 5 | 4 |
| Leeds United | 29.9% | 7 | 6 |
| Liverpool | 26.3% | 4 | 10 |
| Manchester United | 23.3% | 6 | 13 |
Two places are all but taken. The other two are contested by five clubs separated by seven points of probability, which is another way of saying the model cannot split them.
What the model does not see
It reads neither line-ups, nor injuries, nor transfers. It only learns about a changed squad through results, weeks later. For a single fixture we use a second model that starts from the announced eleven — and the two do not always agree.
It refuses to overreact, which is a virtue in September and a flaw in February.
And it gives a probability. City at 59% means they fail in two scenarios out of five.
The same calculation runs every night on six major leagues. Read next: who will win Ligue 1, who will win La Liga, who will win the Bundesliga and who will win the Champions League.
Frequently asked questions
Who will win the Premier League in 2026-27?
Manchester City are champions in 59.5% of the 50,000 simulated seasons run on 20 September 2026, Arsenal in 33.7%. Between them they account for 93% of titles. Brighton are a distant third on 2.8%.
Can Arsenal beat Manchester City over a season?
Yes, in a third of simulations. Arsenal have the best defence in the league, with an index of 0.68 against City's 0.74. The difference is at the other end: 1.51 for City against 1.37 for Arsenal, worth roughly four projected points across a full season.
Can Liverpool still win the title?
The model gives them 0.7%. Tenth on 6 points, Liverpool are the fourth-strongest side in the league: their level is still high, but the deficit to two clearly better clubs is very hard to close over thirty-three matchdays.
Which clubs will be relegated from the Premier League?
Coventry City go down in 83.6% of simulations, Ipswich Town in 56.6%, Tottenham in 47.2% and Crystal Palace in 35.9%. Coventry combine the worst attack in the league with one of its worst defences.
What is this simulated table based on?
A Dixon-Coles model that estimates an attacking and a defensive strength for each club from past fixtures, weighted by how recent they are, then replays the remaining matchdays 50,000 times.
See the table and projections for Premier League →
The figures in this piece are read from the latest published computation: they update with the model and may therefore differ from those of the publication date. That is deliberate — an article should not go on asserting what is no longer true.
