Seven AI Models Split on 2026 World Cup: Spain vs Argentina

Seven advanced AI models were given the full 48-team World Cup draw and asked to forecast a champion. Four models picked Spain and three picked Argentina. All seven placed Spain, Argentina and France in a top tier of contenders.

The models ran as Hermes agents with access to public statistics and were allowed to use different methods. Opus 4.8 Max converted Elo gaps to expected goals with a Dixon-Coles model, ran Monte Carlo simulations and returned Spain at 20 percent. Opus adjusted probabilities for heat, high altitude and long travel and reduced Brazil’s chance to about 8 percent.

GPT 5.5 produced a five-column scorecard that weighted squad quality, tactical control, finishing, player availability and draw. It put Spain at 15–18 percent and projected a narrow Spain win over France in a final. DeepSeek v4 Pro produced a long qualitative dossier and favored Argentina at 18 percent, citing tournament experience and a favorable group, but its report used some outdated coaching information.

Stepfun 3.7 ran 50,000 Elo-based simulations and gave Spain the highest single probability at 33 percent after abandoning an earlier attempt that generated implausible expected-goal numbers. Nemotron 3 Ultra ran a bivariate-Poisson simulation and a separate human-style assessment, producing Spain between 18 percent and 22 percent. MiniMax 2.7 favored Argentina at 18 percent and included a public self-audit that flagged and corrected internal errors; it declined to force a single winner in an evenly matched hypothetical. Qwen 3.5 avoided simulations, separated verified facts from estimates and forecasts, marked its overall confidence as LOW, and used a club-based rating that put Argentina above Spain; it gave Argentina 22 percent and Spain about 10 percent.

The models reported shared concerns that affected forecasts: uncertainty about Erling Haaland’s fitness, Lionel Messi turning 39 during the tournament year, an unpredictable Group D, and the effects of travel, heat and altitude on player performance. Several models flagged low confidence in parts of their input data, documented corrections, or noted outdated information.

The choice of ratings and input data drove many differences. Models that relied on live national-team Elo, where Spain ranks first, tended to favor Spain. Models that leaned on club-based ratings, alternate ranking series or historical pedigree tended to favor Argentina.

Human betting prices differed from many AI outputs. On June 7 the Myriad prediction market placed Spain at 19 percent, France at 17 percent and Argentina at about 10 percent. The tournament begins in days and forecasts from these models include explicit caveats about variance in a 48-team, 104-match format and about the effects of setup choices such as prompts and source selection on results.