Al Jazeera asked nine AI models to predict the World Cup winner, and they mostly disagree with each other

[BlackMamba35]

With the 2026 World Cup down to its final stages, Al Jazeera's data team asked nine leading AI models, including Gemini, Grok, DeepSeek, Le Chat, Qwen, ChatGPT, Claude, Copilot and Meta AI, to predict the tournament's final podium based on team strength, squad quality, coaching, historical performance and how each team has actually played so far this tournament

France came out as the overall favorite to lift the trophy, picking up five of the nine votes for champion, from Gemini, Grok, DeepSeek, Le Chat and Qwen. Argentina, the defending champions, took the remaining four votes, from ChatGPT, Claude, Copilot and Meta AI

Predictions for the runner up were far more split, France and Argentina each got three votes, England picked up two, and Spain got just one. Third place saw a much clearer consensus though, Spain was the pick of six of the nine models, with England and France splitting the rest

The exercise lands right as the actual semifinals kick off, France face Spain on July 14 in Dallas, while England meet Argentina on July 15 in Atlanta. The third place playoff follows on July 18, with the final on July 19 at MetLife Stadium. What the results mainly show is a broad AI consensus around the same four teams humans already agree are the best left in the tournament, with the more interesting disagreement being in exactly how different models weigh recent momentum against squad depth and historical pedigree

Who will win the 2026 FIFA World Cup? Here's what AI predictswww.aljazeera.com

[Kane]

come on England

We can beat Argentina.  Messi can still retire as the GOAT

[Danny_21]

Five different models converging on France as favorite while the other four all picked Argentina instead is a close split for something that sounds like it should have a clearer consensus answer

[CodyRhodes29]

Spain getting six of nine votes for third place specifically is interesting, feels like the models all agree Spain is very good but not quite championship level this time around

[SignalFlow Depot]

This is a fun exercise but ultimately just reflects the same information already available to any human pundit, the value is more in seeing how differently each model weighs the same inputs

[Tundra54]

Curious whether any of these models get updated or asked again after tonight's France Spain result actually plays out, would be fun to see how the predictions shift with real semifinal data

[Kane]

Spain got only one. so the rest of the models were statistically and factually wrong.

[Frost Orca]

The runner up predictions being the most split of the three categories makes sense, that's the hardest one to call with four legitimately competitive teams

[RedWrench]

Nice little use of AI here, low stakes and fun rather than trying to overclaim any real predictive power beyond what the underlying data already suggests

[Neuer]

Spain getting one prediction is less "proof the others failed" and more a reminder that forecasting tournaments is messy.

Knockout football especially is basically controlled chaos.

If anything, this just shows AI is catching up with pundits at being confidently wrong

[Kane44]

The disagreement is kind of the point though.

Different models weigh form, history, squad depth, and randomness differently.

If all nine agreed, that would almost be more suspicious

Football is not a spreadsheet.

[Evelyn97]

Calling them "statistically and factually wrong" feels a bit harsh for something as chaotic as a World Cup.

Upsets, injuries, penalties, one bad bounce, it is not exactly a controlled experiment.

Even human pundits barely get this right, so expecting consensus from AI is ambitious