WorldCupArena: Fine-Grained Evaluation of Language Models and Deep-Research Agents on Football Forecasting
The benchmark tests whether models can make football predictions before results exist, then grades them after the match.
WorldCupArena starts with the 2026 FIFA World Cup and is designed to be reused for future competitions. Models either get a shared evidence pack or search on their own, then predict results, scores, players, events, stats, and tournament outcomes. The paper says similar win-result accuracy can hide bigger gaps in detailed forecasting, especially scoreline quality. The best system only slightly beat betting-market and fan baselines on result and exact-score accuracy, but did better on the Scoreline metric. HF Daily Papers' note
WorldCupArena starts with the 2026 FIFA World Cup and is designed to be reused for future competitions. Models either get a shared evidence pack or search on their own, then predict results, scores, players, events, stats, and tournament outcomes. The paper says similar win-result accuracy can hide bigger gaps in detailed forecasting, especially scoreline quality. The best system only slightly beat betting-market and fan baselines on result and exact-score accuracy, but did better on the Scoreline metric. HF Daily Papers' note
score 4