Experiment 01 · Oracle
One shot. Twenty oracles.
Each model predicts the full FIFA World Cup 2026 bracket once, before the opening match. Below is what every model called — champion, runner-up, third place, top scorer — and how accurate those calls turned out to be over all 104 matches.
Champion
Spain
Runner-up
Argentina
Third
England
Top scorer
Kylian Mbappe
| # | Model | Accuracy | Champion | Runner-up | Third | Top scorer |
|---|---|---|---|---|---|---|
Llama 4 Maverick | 38.0% | BRA | ARG | FRA | Vinicius Jr8 | |
Qwen 3.7 Plus | 38.0% | FRA | ARG | FRA | Vinicius Jr8 | |
GPT-5 Mini | 37.5% | ARG | ESP | FRA | Vinicius Jr8 | |
04 | Claude Haiku 4.5 | 35.1% | BRA | ARG | FRA | Vinicius Jr8 |
05 | Gemini 2.5 Pro | 35.1% | BRA | ARG | FRA | Kylian Mbappe10 |
06 | Claude Fable 5 | 35.1% | ESP | ARG | FRA | Vinicius Jr8 |
07 | Claude Opus 4.8 | 34.6% | BRA | ESP | FRA | Vinicius Jr8 |
08 | Gemini 2.5 Flash | 34.1% | BRA | ESP | FRA | Kylian Mbappe12 |
09 | Moonshot Kimi K2.6 | 34.1% | BRA | ARG | FRA | Erling Haaland9 |
10 | Llama 3.3 70B | 34.1% | ARG | ESP | FRA | Christian Pulisic8 |
11 | Mistral Large | 34.1% | BRA | ARG | FRA | Lamine Yamal8 |
12 | Perplexity Sonar | 34.1% | FRA | ESP | FRA | Vinicius Jr8 |
13 | Claude Sonnet 4.6 | 33.7% | ARG | ESP | FRA | Lionel Messi9 |
14 | GPT-5.5 | 32.7% | ESP | ESP | FRA | Vinicius Jr8 |
15 | Qwen 3.7 Max | 32.7% | BRA | ESP | FRA | Kylian Mbappe9 |
16 | DeepSeek V4 Pro | 32.7% | ARG | ARG | FRA | Vinicius Jr8 |
17 | Grok 4.20 | 32.2% | BRA | ARG | FRA | Erling Haaland9 |
18 | GPT-5 Nano | 31.7% | BRA | ARG | FRA | Santiago Gimenez5 |
19 | Mixtral 8x22B | 31.3% | FRA | ARG | FRA | Vinicius Jr8 |
20 | DeepSeek R1 0528 | 31.3% | ESP | ESP | FRA | Vinicius Jr7 |
21 | Gemini 2.5 Flash Lite | 30.3% | BRA | ESP | FRA | Vinicius Jr9 |