Summary

  • Seven advanced AI models were tasked with predicting the winner of the 2026 World Cup based on the same tournament draw.
  • Four models favored Spain, with Stepfun estimating a 33% chance; three models supported Argentina, with Qwen estimating a 22% chance. Spain, Argentina, and France were consistently rated at the top.
  • Myriad, a prediction market owned by Decrypt's parent company, aligned with these models, ranking Spain at 19%, France at 17%, and Argentina at just 10% as of June 7.

The 2026 World Cup is set to begin shortly, prompting widespread speculation about its outcome.

Everyone has their predictions, from friends and colleagues to the latest AI technology.

Artificial intelligence has become a trusted resource for various tasks, including predicting sports outcomes. These models analyze team statistics, performance trends, and more to provide their forecasts with a level of confidence that humans can only simulate.

This isn't the first time I've sought predictions from AI; I previously consulted an AI team for March Madness (which did not fare well) and a customized HorseGPT for the Kentucky Derby (which performed surprisingly well). It's a mix of useful insights and humbling experiences.

With the biggest global tournament approaching, we decided to conduct a larger analysis this time.

We developed Hermes agents and provided them access to various statistics websites, equipping them with unique skills to predict the World Cup champion and rank all participating teams. Each model was given the tournament's complete details, including 48 teams and 12 groups, and was free to choose its analytical approach.

We then observed how they arrived at their conclusions.

Four models selected Spain as the winner, while three favored Argentina. The differences in their predictions stemmed from the data each model prioritized.

Here’s a breakdown of the predictions made by each AI.

Opus 4.8 Max — The Meteorologist

Prediction: Spain. 20% / Dixon-Coles Poisson + Monte-Carlo simulations · Final: Spain defeats France

Opus 4.8 Max analyzed the World Cup as if it were a physics equation. It utilized Elo ratings to calculate expected goals, employing a Dixon-Coles model and simulating the tournament thousands of times. It determined that Spain had a 20% chance of winning against France in the final, with Portugal and England eliminated in the semifinals.

Its analysis included many off-pitch factors, such as environmental conditions, which other models often overlook. Opus highlighted that certain matches would be played in extreme heat that could affect players' performance, and noted that teams traveling to high altitudes might struggle.

It significantly reduced Brazil's odds to 8%, citing injuries to key players like Rodrygo and Neymar. Its boldest prediction was Spain defeating Argentina in the quarterfinals, where Messi's performance would be critical.

GPT 5.5 — The Careful Scout

Prediction: Spain 15–18% / Five weighted categories, no simulations · Final: Spain 2-1 France

OpenAI's GPT 5.5 opted for a scorecard approach instead of relying on large statistics. It evaluated teams based on five weighted factors, with squad quality being the most significant. It assigned Spain a 15–18% chance of winning and projected a close final score of 2-1 against France.

It validated its findings against a supercomputer and consulted Spanish sports media for insights that might not be captured by models. While it noted some minor issues in the Spanish squad, such as injuries, these did not alter its prediction.

DeepSeek v4 Pro — The Maximalist

Prediction: Argentina 18% / Detailed qualitative analysis · Final: Argentina vs France

DeepSeek v4 Pro produced an extensive analysis, detailing all teams and their travel distances. It ultimately favored Argentina, with an 18% chance of winning against France in the final, despite an error in the location of the final match.

It argued that Argentina's calmness and coaching strategy would help manage Messi's performance effectively. However, it relied on outdated coaching information, which could have affected its conclusions.

Stepfun 3.7 — The True Believer

Prediction: Spain 33% / Pure-Elo Monte Carlo, 50,000 simulations · Final: Spain vs Argentina

Stepfun 3.7 was the most confident model, running 50,000 simulations and giving Spain a striking 33% chance of winning, with Argentina at 15%. It initially struggled with an overly complicated prediction model but then reverted to a simpler Elo-based approach. While it provided a direct prediction, it did not factor in human elements like player injuries or external conditions.

Nemotron 3 Ultra — The Double-Checker

Prediction: Spain 18–22% / Bivariate Poisson + subjective assessment · Final: Spain vs Argentina

Nvidia's Nemotron 3 Ultra ran the tournament twice, first through simulations and then with a manual evaluation of teams. Both methods resulted in Spain being favored, with odds of 18% and 22%. It provided extensive insights into team tactics and formations, even predicting Türkiye would win its group.

MiniMax 2.7 — The Self-Auditor

Prediction: Argentina 18% / Qualitative, self-assessment · Final: Argentina vs France, no scoreline

MiniMax 2.7 estimated Argentina's chances at 18%, just ahead of France, and engaged in self-critique throughout its report. It corrected previous errors and avoided making assumptions about match outcomes.

Qwen 3.5 — The Contrarian with Evidence

Prediction: Argentina 22% / Research-focused, no simulations · Final: Argentina 2-1 Spain

Qwen 3.5, a 397-billion-parameter model, focused solely on factual evidence and did not run simulations. It predicted Argentina would defeat Spain 2-1, but it placed Spain in fifth with a low 10% chance of winning. Its reasoning emphasized the importance of teamwork and strategy over individual talent.

Commonalities in Predictions

Despite differing predictions, all seven models included Spain, Argentina, and France among their top contenders and identified similar group winners. The main divergence was based on the data sources used, leading to differing conclusions about the champion.

Market Sentiment

In the betting market, Spain is currently the favorite at 19%, followed closely by France at 17%, with Argentina trailing at just 10% as per Myriad.

The human bettors appear to be more cautious about Argentina than the AI models, reflecting a more conservative outlook.

Conclusion

While these predictions are insightful, it’s essential to remember that they are not definitive. Even the most confident models recognize the uncertainty inherent in sports outcomes. With 48 teams and numerous matches ahead, the unpredictability of football remains intact.

Ultimately, these AI predictions provide a starting point for discussion rather than a reason to place significant bets. Four models predict Spain as the winner, while three favor Argentina, but the outcome will ultimately be decided on the pitch.