WorldCup Arena: leaderboard
WorldCup Arena (2026) evaluates frontier LLMs prospectively on the 2026 FIFA World Cup: before each of the 104 kickoffs a model filled a seven-market prediction card, and before the tournament it picked the 12 group winners and an outright pool, so no answer existed when the question was asked. Total points out of 1,790 under weights fixed before the tournament; higher is better. Every model ran at its highest thinking setting with its provider-native web search and identical team dossiers.
Metric: Total points (out of 1790); a seven-market prediction card for each of the 104 matches worth up to 16 points (handicap 4, half-time/full-time 3, match outcome, over/under 2.5 goals, both teams to score and correct score 2 each, odd/even goals 1), 12 group winners at 5 points and a 66-point pre-tournament outright pool; predictions made live before each kickoff of the 2026 FIFA World Cup (10 June to 19 July 2026) from byte-identical rolling team dossiers, each model at the highest thinking setting its API exposes and with its provider's own server-side web search. Source: arxiv.org. Saturation forecast: Around August 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 (Thinking) | 897 | #56 (Claude Opus 4.8) |
| 2 | Seed 2.0 Pro (High) | 853 | #209 (Seed 2.0 Pro) |
| 3 | GPT-5.5 (High) | 845 | #48 (GPT-5.5) |
| 4 | Kimi K2.6 | 839 | #173 |
| 5 | Gemini 3.1 Pro (Preview) | 827 | #91 |
| 6 | GLM-5.2 | 813 | #137 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=worldcup-arena · How It Works · Data refreshed daily, snapshot 2026-11-12.