The biggest football tournament of 2026, forecasted by TabPFN
Every match predicted by our tabular foundation model. We simulated the tournament thousands of times to find the most likely champion. Want to build your own predictive model with TabPFN? Read our methodology.
Updated 18 Jun 2026 · awaiting new data
Matches
ColombiaDR Congo
Picks locked — the match day has started
69%21%10%
Your pick
EnglandGhana
Picks locked — the match day has started
86%11%3%
Your pick
PanamaCroatia
Picks locked — the match day has started
20%26%54%
Your pick
PortugalUzbekistan
Picks locked — the match day has started
64%23%12%
Your pick
Groups
Overall Winner
Competition
Prediction methodology
We used a dataset of historical national games to generate game-outcome predictions with TabPFN-3. On top of the raw data, we engineered features like Elo rating, form and recent goal differential that take short-term information into account.
Feature engineering
The feature set is built in a single chronological pass over the data, so every feature at kickoff time uses only matches that have already been played — no future leakage.
- Elo ratings
elo_diffhome_eloaway_elo - The most important features. Each team starts at 1500 and is updated after every result using the standard formula, scaled by two multipliers: margin of victory (a 3-goal win counts more than a 1-goal win) and tournament importance — major-tournament matches move the needle more than friendlies. The raw ratings capture long-run strength; the difference between the two teams, with home advantage folded in as a 65-point bonus for non-neutral venues, is the clearest single signal the model has.
- Form
form5_diffform10_diffhome_form5away_form5 - Average points-per-game over the last 5 and 10 matches. Two windows, because a team can be tactically hot over 5 games while a 10-game window catches a longer drift. The differential between teams compresses both into one number without losing the individual values.
- Goal stats
home_gf5away_gf5home_ga5away_ga5gd10_diff - These go beyond results. A team winning 1-0 every week and one winning 4-1 carry the same points tally but very different attacking profiles. Average goals scored and conceded over the last 5 games, combined with a 10-game goal differential, let the model distinguish them.
- Streak and rest
home_streakaway_streakhome_restaway_rest - These round out the short-term picture. Consecutive wins capture momentum that points-per-game smooths over. Days since the last game (capped at 90) capture fatigue from fixture congestion, which matters in tournament group stages.
- Head-to-head history
h2h_nh2h_home_winrateh2h_draw_rateh2h_gd - Captures matchup-specific dynamics that aggregate ratings miss. Some pairings have persistent tactical or psychological patterns that show up reliably over dozens of meetings.