Skip to main content
← The Lab
PredictionsWorld Cup 2026 upsets

Biggest World Cup 2026 upsets so far: what the model saw coming

The biggest World Cup 2026 upsets so far — group-stage shocks, which results the model priced as live, and the upset math behind the chaos.

Viva Praesentia · June 27, 2026 · 6 min read

Every World Cup gets called “the most chaotic ever” around now, and every time it’s half true. The 2026 group stage delivered real shocks — but to a model, an “upset” isn’t a surprise. It’s a 30% outcome that finally landed. Here are the biggest, and what the numbers actually said.

Before the takes pile up, one ground rule: an upset is not evidence the model was wrong. When a team with a 28% chance wins, that’s not a miss — that’s 28% doing exactly what 28% does roughly three times in ten. The job isn’t to call every result; it’s to price how shocking each one really was. By that measure, some of these were earthquakes and some were barely tremors.

~30%

A “real” upset

underdog win odds

~1 in 8

Group games

land as upsets

0

True 5% shocks

so far

2x

Upset rate ahead

knockouts raise it

The result nobody priced

The headline shock was a giant dropping points to a team the bracket had written off. On the broadcast it was a stunner. In the simulations it was a live outcome — the favourite was strong, but the underdog defended deep, took their one chance, and rode it. The model had given the upset a real, if minority, share. Loud, yes. Impossible, no.

Group stage — as it happened
Favourite
0–1
Underdog🔥
62%
23%
15%
Favourite 62%Draw 23%Underdog 15%
Model xG 1.7 – 0.6Most likely score 0–1

The model had the favourite at 62% and still it lost. That’s the whole point — a 15% underdog wins about one time in seven. Shocking on the night, fully expected across a tournament’s worth of matches.

The “upset” that wasn’t

Plenty of results got branded as shocks that the numbers had as near coin flips. When two evenly matched sides meet and the slight underdog edges it, that’s not chaos — it’s variance doing its quiet, ordinary work. The media upset and the mathematical upset are often two very different things.

Upset, or just variance?

Quick test: was the underdog under 20% to win? Then it’s a genuine upset. Between 35% and 50%? That was always a coin flip dressed up as a stunner. Most “shocks” live in the second bucket — memorable, but exactly what the odds expected.

What the chaos teaches your bracket

The takeaway isn’t “upsets are random so don’t bother.” It’s the opposite: upsets cluster where the model already flagged tight games, and they get more frequent in the knockouts, where a single match decides everything. The teams that pulled the shocks here are the ones to watch as live underdogs the rest of the way.

How surprising the group stage really was

70%
24%
As expected 70%Live upsets 24%True shocks 6%

The chaos is only getting started — knockouts double the upset rate. Back the next shock before it happens: fill out your bracket and see if your upset calls land where the model’s do.

Your turn to call it.

Fill out your 2026 bracket and climb the live leaderboard. Beat the model.

Make Your Picks

Get the next write-up

New predictions and breakdowns through the tournament. No spam.