All four models have submitted a full 48-team forecast from the same context pack. The next update lands after the group stage. Last updated: 8 June 2026.
The 2026 World Cup starts this week. There are 48 teams spread across the United States, Canada, and Mexico, a bigger group stage than we’re used to, and a knockout bracket that goes all the way to 19 July.
Before anyone kicks a ball, I want to know what the big language models think is going to happen. I also want their picks written down somewhere, so that once it’s all over I can hold them up against what really happened and see who looks silly.
So that’s what this page is for. Right now it has the pre-tournament forecasts. After each round I’ll add a new dated section and score the models on how they did, and when the final is done I’ll write up a post-mortem.
Table of contents
Open Table of contents
What I gave them
I used four models and scored each one separately:
- Claude Opus 4.7 (Anthropic)
- Claude Sonnet 4.6 (Anthropic)
- GPT-4o (OpenAI)
- Gemini 2.5 Pro (Google)
Every model got the exact same context pack. It had 48 team dossiers, the official Group A to L draw, squad ratings, manager bios, recent qualifying form, injury news, betting-market odds, and head-to-head history. I made sure there were no predictions anywhere in the pack.
I kept the setup strict. No web access, no tools, and no live data. Each model got the same instruction and sent back a structured forecast. In that forecast it gave a champion probability for all 48 teams, plus a probability for each team reaching each stage (group winner, round of 16, quarter, semi, final, champion), plus its single most likely winner for every group.
Then I did a second, separate run. This time each model guessed the exact score of all 72 group-stage matches. From that I get full standings for every group and a match-by-match card to go with them.
Every group's predicted standings and all 72 match scorelines from the four models, side by side: see the full group-stage predictions →
Where they agreed
Spain is the favourite on all four models. The average champion probability for Spain comes out at 15.2%, and none of the models gives its top team more than 16.0%. I think that’s about right for an open 48-team tournament. If a forecast is well calibrated it shouldn’t push any one team much past 18 to 20%, and all four models sit comfortably under that.
Below is the top 12, sorted by the average champion probability, with each model’s own number next to it:
| Rank | Team | Opus 4.7 | Sonnet 4.6 | GPT-4o | Gemini 2.5 Pro | Mean |
|---|---|---|---|---|---|---|
| 1 | Spain | 15.5% | 16.0% | 13.8% | 15.5% | 15.2% |
| 2 | France | 14.5% | 14.9% | 12.1% | 14.5% | 14.0% |
| 3 | England | 11.5% | 11.9% | 11.6% | 13.4% | 12.1% |
| 4 | Argentina | 10.5% | 10.8% | 11.2% | 10.3% | 10.7% |
| 5 | Brazil | 9.5% | 8.8% | 9.9% | 7.2% | 8.9% |
| 6 | Portugal | 7.0% | 6.7% | 7.3% | 8.2% | 7.3% |
| 7 | Germany | 6.0% | 9.3% | 6.0% | 6.2% | 6.9% |
| 8 | Netherlands | 5.0% | 4.6% | 3.9% | 5.1% | 4.7% |
| 9 | Belgium | 3.0% | 2.6% | 3.0% | 2.1% | 2.7% |
| 10 | Colombia | 1.8% | 1.9% | 1.7% | 2.6% | 2.0% |
| 11 | Uruguay | 1.5% | 1.5% | 1.3% | 1.5% | 1.5% |
| 12 | Croatia | 1.5% | 0.9% | 1.7% | 1.5% | 1.4% |
Every forecast has the same six teams at the top: Spain, France, England, Argentina, Brazil, Portugal. On each model, those six add up to roughly two-thirds of the total chance of winning the whole thing.
On group winners the models lined up even more closely. Here’s the pick each model rated most likely in each group:
| Group | Opus 4.7 | Sonnet 4.6 | GPT-4o | Gemini 2.5 Pro |
|---|---|---|---|---|
| A | Mexico (52%) | Mexico (52%) | Mexico (58%) | Mexico (55%) |
| B | Switzerland (54%) | Switzerland (56%) | Switzerland (57%) | Switzerland (58%) |
| C | Brazil (74%) | Brazil (78%) | Brazil (82%) | Brazil (82%) |
| D | USA (42%) | Turkey (38%) | USA (46%) | Turkey (38%) |
| E | Germany (62%) | Germany (68%) | Germany (67%) | Germany (70%) |
| F | Netherlands (52%) | Netherlands (62%) | Netherlands (60%) | Netherlands (65%) |
| G | Belgium (58%) | Belgium (58%) | Belgium (65%) | Belgium (72%) |
| H | Spain (74%) | Spain (80%) | Spain (81%) | Spain (80%) |
| I | France (66%) | France (78%) | France (78%) | France (78%) |
| J | Argentina (72%) | Argentina (78%) | Argentina (80%) | Argentina (75%) |
| K | Portugal (62%) | Portugal (65%) | Portugal (70%) | Portugal (68%) |
| L | England (70%) | England (72%) | England (74%) | England (76%) |
In eleven of the twelve groups, all four models picked the same most likely winner. Group D is the one place where they split.
You see the same thing match by match. Each model also gave a probability for every one of the 72 group-stage games, and in 69 of those all four went for the same most likely result.
Where they disagreed
There were two splits worth talking about.
Group D: USA vs Türkiye. Opus and GPT-4o hand the group to the USA, who are hosting (42% and 46%). Sonnet and Gemini go with Türkiye instead (38% each). According to the pack, Türkiye has the stronger squad on average, built around Çalhanoğlu, Güler, and Yıldız. The USA, on the other hand, has Pochettino and a home crowd behind it. So two of the models leaned on squad quality, and the other two gave more weight to playing at home.
Brazil vs England as the third elite. GPT-4o has England third on 11.6%, just ahead of Brazil on 9.9%. Gemini also has England third, at 13.4%, but drops Brazil all the way to sixth at 7.2%, which is the biggest gap anywhere in the panel. Opus and Sonnet keep Brazil close, though still a little behind England. Most of this split comes from Gemini, which likes England more than any other model does and is also the most doubtful about Brazil.
How I will score them
I’m going to score two separate things, and it’s easy to mix them up.
The first one is agreement, and I can already score that. All four models agree on 11 of the 12 group winners and on 69 of the 72 group-stage match results, and they all put Spain first, somewhere between 13.8% and 16.0%. That tells me the models share a view. It doesn’t tell me the view is correct.
The second one is accuracy, which I’ll score as the games get played. The winner, runner-up, Golden Boot, and third place all get settled at the final. The group winners get settled once the group stage ends. Having probabilities for all 72 matches means I can score each model on a much bigger sample than just the headline picks, and honestly that’s the part I care about most.
I’d also like to look at calibration. When a model says Spain has a 15.5% chance of winning, can I trust that number? One tournament can’t fully answer that. But if I check the 72 group-stage match probabilities against the real results, I get a rough reliability check for each model, and I expect that to be the most useful thing to come out of all this.
Live updates
Here’s what I plan to do after each round:
- Group stage: score the 72 match results, score each model on its full set of match probabilities, and point out any group winners the consensus got wrong, Group D included.
- Round of 16: call out any model whose knockout bracket is already busted, and score the round-of-16 probabilities.
- Quarter-finals: keep track of which model still has the most picks alive.
- Semis and final: this is where the winner and third place get decided. I’ll score each model’s champion forecast against whoever actually lifts the trophy.
- Post-mortem: the full scorecard. I’ll cover what each model got right, where the consensus missed by the most, and whether the four models together did better than the betting market.
The pre-tournament forecasts are up today. Expect the first round-of-16 update not long after 27 June.