Can AI call the beautiful game? 44 LLMs — flagships to a 1B miniature, 2026 frontier to 2023 legends — predicted the entire World Cup before the opening kickoff.

Tournament progress: 0 of 104 matches played

Champion board

Every model simulated its own tournament to the end — these are their champions.

Netherlands ×2

DeepSeek V4 Flash · Hunyuan A13B

Mexico ×1

GPT-5.4 Nano

4 models without a valid bracket — couldn't produce valid knockout predictions within the retry policy (see methodology)

Leaderboard

#ModelTotalGroup ptsBracket ptsExactChampion pick
1
Claude 3 HaikuAnthropiclegacy
0000 Brazil
1
Claude Fable 5Anthropicflagship
0000 Argentina
1
Claude Haiku 4.5Anthropicsmall
0000 Germany
1
Claude Opus 4.8Anthropicmid
0000 Brazil
1
Command ACohereflagship
0000 France
1
DeepSeek V4 FlashDeepSeeksmall
0000 Netherlands
1
DeepSeek V4 ProDeepSeekflagship
0000 Argentina
10000 Argentina
10000 Brazil
10000 Brazil
1
Gemma 2 27BGooglelegacy
0000 France
1
GLM 4.7 FlashZ.AIsmall
0000 Germany
1
GLM 5.1Z.AIflagship
0000 Germany
1
GPT-3.5 TurboOpenAIlegacy
0000 France
1
GPT-4OpenAIlegacy
0000 France
1
GPT-4oOpenAIlegacy
0000 Argentina
1
GPT-5.4 MiniOpenAIsmall
0000 Germany
1
GPT-5.4 NanoOpenAIsmall
0000 Mexico
1
GPT-5.5OpenAIflagship
0000 Argentina
1
GPT-5.5 ProOpenAIflagship
0000 Spain
10000no valid bracket
1
Grok 4.20xAImid
0000 Brazil
1
Grok 4.3xAIflagship
0000 Brazil
1
Hermes 3 405BNous Researchoddball
0000 Germany
1
Hunyuan A13BTencentoddball
0000 Netherlands
1
Jamba Large 1.7AI21flagship
0000 Brazil
1
Kimi K2.6Moonshotflagship
0000 Germany
1
LFM-2 24BLiquid AIoddball
0000no valid bracket
1
Llama 3 70BMetalegacy
0000 Brazil
1
Llama 3.2 1BMetaoddballpredictions pending
0000
1
Llama 4 MaverickMetaflagship
0000 Spain
1
Llama 4 ScoutMetasmall
0000 Brazil
1
Mercury 2Inceptionoddball
0000 Spain
1
MiniMax M3MiniMaxflagship
0000 Spain
1
Mistral Medium 3.5Mistralflagship
0000 Germany
1
Mistral Small 4Mistralsmall
0000 Argentina
1
Nemotron 3 UltraNVIDIAflagship
0000 France
1
Nova 2 LiteAmazonmid
0000 Germany
1
Phi-4 MiniMicrosoftsmall
0000no valid bracket
1
Qwen 2.5 72BAlibabalegacy
0000 Germany
1
Qwen3.6 FlashAlibabasmall
0000 Brazil
1
Qwen3.7 MaxAlibabaflagship
0000 France
1
Qwen3.7 PlusAlibabamid
0000 Argentina
1
WizardLM-2 8x22BMicrosoftoddball
0000 France

Total = group match points (exact 3 · goal difference 2 · outcome 1) + bracket points: advancement for every real team a model had reaching each stage (R32 1 · R16 2 · QF 3 · SF 5 · final 8 · champion 13), +1 per simulated pairing that actually occurs, and matched pairings' scorelines scored like normal matches. Bracket points pay out once the real knockout bracket forms. Tiebreakers: points → exact scores → correct champion → correct R32 qualifiers.

Next matches

What is this?

Before the opening kickoff, 33 large language models each predicted the entire 2026 World Cup — every group-stage score and, derived from those scores, their own group tables, their own knockout bracket and their own champion. Reality then grades every claim: group matches on exact score, goal difference and outcome; brackets on which real teams a model had reaching each stage, on simulated pairings that actually happen, and on the scorelines it attached to them. Everything was locked and SHA-256 pre-registered before the first match, so nothing can be quietly edited after the fact.

Read the full methodology →