Seven frontier AI models were given the same task: forecast the 2026 World Cup champion. The result was a split decision. Four models picked Spain, three chose Argentina. Every model, however, placed Spain, Argentina, and France in their top tier.
The division wasn't about football intuition. It was about data. Models using live football Elo ratings favored Spain. Those relying on FIFA rankings or 2022 pedigree leaned toward Argentina.
Anthropic's Opus 4.8 Max treated the tournament like a physics problem, simulating thousands of brackets. It picked Spain at 20%, citing heat and altitude as factors that favor deep European squads. It gutted Brazil's odds to 8%.
OpenAI's GPT 5.5 built a scorecard across five weighted columns. It also chose Spain, but with a cautious 15-18% probability, emphasizing "ranges rather than fake precision."
Nvidia's Nemotron 3 Ultra ran the tournament twice-once with cold simulation, once with human-style subjective scoring. Both versions crowned Spain.
Stepfun 3.7 was the most confident, picking Spain at 33% after 50,000 simulated tournaments. It was also the most transparent about its failures.
On the other side, DeepSeek v4 Pro and MiniMax 2.7 both picked Argentina at 18%, citing the champions' calm spine and a soft group. Qwen 3.5 was the biggest rebel, picking Argentina at 22% and dropping Spain to just 10%.
Prediction markets agree on Spain as the favorite at 19%, with France at 17%. Argentina sits at just 10%.
Ultimately, the models proved that even advanced AI struggles with football's unpredictability. As one model put it, the beautiful game will do exactly as it pleases.