2026-fifa-world-cup-d65f4e83·1 events·first seen Aliases: 2026 FIFA World Cup
WorldCupArena is a new dynamic benchmark that evaluates language models and deep-research agents on football match prediction tasks using the 2026 FIFA World Cup as its first evaluation corpus. Models predict match results, scores, player events, and competition outcomes either from a provided evidence package or via autonomous web search, with predictions scored post-match against actual results across 104 matches and 13 systems. The benchmark finds that models with similar result accuracy diverge more clearly on fine-grained predictions, and that the best system shows only modest gains over betting-market and human-fan baselines. The benchmark is designed to be reusable for future tournaments, avoiding data contamination by construction.