We Asked 7 AI Agents to Predict the 2026 World Cup: Here's What They Said
Decrypt asked seven AI models to forecast the 2026 World Cup. Four picked Spain, three picked Argentina, and Myriad's market also leans Spain.
Intelligence analysis by GPT-5.4 Mini
Decrypt ran seven frontier AI models against the same 2026 World Cup draw and compared their picks with Myriad's market. Spain and Argentina dominated the forecasts, but the article treats the exercise as a test of how models reason, not a true prediction.
A bunch of smart robot guessers were asked to pick the next World Cup winner, like kids filling out a tournament bracket. Most picked Spain or Argentina, but the story says the robots are still only making educated guesses, not magic predictions.
Analysis
What Decrypt did
Decrypt set up Hermes agents with access to statistics sites and custom skills, then gave seven AI models the same task: predict the 2026 World Cup champion and explain the path there. The models were fed the real 48-team draw and could use their own approach.
Where the models landed
The forecasts split mainly between Spain and Argentina. Four models chose Spain, including one that leaned heavily on simulation and another that used Elo ratings and Monte Carlo runs. Three models chose Argentina, often arguing that defending champions, squad continuity, or tournament experience mattered most. Across the board, Spain, Argentina, and France formed the clear top tier.
How they differed
The article is less about one correct answer than about the methods. Some models simulated thousands of brackets, some built weighted scorecards, some wrote long qualitative reports, and one even self-audited its own uncertainty. The strongest Spain pick reached 33%, while the strongest Argentina pick reached 22%.
Market check
Decrypt compares the AI output with Myriad, the prediction market run by its parent company Dastan. As of June 7, Myriad had Spain at 19%, France at 17%, and Argentina at 10%, which put the market closer to the Spain side than the Argentina side.
The caution
The article closes by stressing that none of this is a crystal ball. It notes that even strong football models are only right a bit more than half the time, and that the larger 48-team format adds more uncertainty. It also flags a basic limitation: the models can inherit bias from their prompts, their data, and the humans who set them up.
Key points
- Decrypt asked seven frontier AI models to forecast the 2026 World Cup using the same draw and supporting data.
- Four models picked Spain as champion, while three picked defending champion Argentina.
- Every model placed Spain, Argentina, and France in its top tier of contenders.
- Myriad, Decrypt's parent company's prediction market, also favored Spain, with France next and Argentina lower.
- The article argues the exercise is useful for testing reasoning, but not reliable enough to treat as a real crystal ball.
If AI forecasting keeps improving, it could become a useful way to compare teams, surface hidden variables, and complement human judgment. The article shows some models already cross-checking ratings, injuries, and market signals instead of guessing blindly.
The piece also shows how easily models can drift into hallucinations, outdated data, or prompt bias. If the setup is wrong, even a confident forecast can mislead, especially in a 48-team tournament with many unpredictable matches.
