Experiment 01 · Oracle
One shot. Twenty-one 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 — scored by trajectory skill: how closely the tournament it imagined tracked the one that actually happened. Because each model plays its own bracket, there is no shared fixture list to grade its knockout scorelines against.
Champion
Spain
Runner-up
Argentina
Third
England
Top scorer
Kylian Mbappe
| # | Model | Trajectory skill | Champion | Runner-up | Third | Top scorer |
|---|---|---|---|---|---|---|
GPT-5.5 | 49.8% | ESP | FRA | BRA | Alexander Isak8 | |
DeepSeek R1 0528 | 47.8% | ESP | FRA | BRA | Kylian Mbappe7 | |
Claude Fable 5 | 47.8% | ESP | FRA | BRA | Kylian Mbappe8 | |
04 | Llama 3.3 70B | 47.0% | ARG | ESP | USA | Christian Pulisic8 |
05 | Claude Sonnet 4.6 | 45.8% | ARG | BRA | ESP | Lionel Messi9 |
06 | GPT-5 Mini | 45.3% | ARG | BRA | ESP | Enner Valencia8 |
07 | Grok 4.20 | 44.9% | BRA | ARG | ESP | Erling Haaland9 |
08 | DeepSeek V4 Pro | 42.0% | ARG | BRA | ESP | Lamine Yamal8 |
09 | Qwen 3.7 Max | 31.7% | BRA | FRA | ESP | Kylian Mbappe9 |
10 | Mistral Large | 31.2% | BRA | FRA | ESP | Kylian Mbappe8 |
11 | Qwen 3.7 Plus | 28.1% | FRA | BRA | GER | Kylian Mbappe8 |
12 | Perplexity Sonar | 24.9% | FRA | GER | BRA | Kylian Mbappe8 |
13 | Gemini 2.5 Pro | 24.8% | BRA | FRA | GER | Kylian Mbappe10 |
14 | Moonshot Kimi K2.6 | 24.6% | BRA | FRA | GER | Erling Haaland9 |
15 | Gemini 2.5 Flash | 24.4% | BRA | FRA | GER | Kylian Mbappe12 |
16 | Claude Haiku 4.5 | 23.1% | BRA | FRA | GER | Kai Havertz8 |
17 | Claude Opus 4.8 | 22.2% | BRA | FRA | GER | Kenan Yildiz8 |
18 | Llama 4 Maverick | 18.1% | BRA | FRA | GER | Kylian Mbappe8 |
19 | Gemini 2.5 Flash Lite | 17.3% | BRA | FRA | BEL | Kylian Mbappe9 |
20 | Mixtral 8x22B | 17.1% | FRA | BRA | GER | Kylian Mbappe8 |
21 | GPT-5 Nano | 17.0% | BRA | FRA | NED | Raul Jimenez5 |