FanOutQA: leaderboard
FanOutQA — multi-hop 'fan-out' question answering over Wikipedia that requires aggregating facts about many entities into one answer.
Metric: Accuracy (%, closedbook loose). Source: fanoutqa.com. Saturation forecast: Around 2030. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mistral Large 2 (Jul) | 47.45 | #484 |
| 2 | Mixtral 8x7B Instruct | 46.96 | #1170 |
| 3 | Llama 3 70B Instruct | 46.62 | #634 |
| 4 | GPT-OSS-120B | 46.6 | #341 |
| 5 | GPT-4 Turbo | 45.97 | #404 |
| 6 | Mistral-Small-Instruct-2409 | 45.01 | #702 |
| 7 | Claude 3 Opus | 44.77 | #571 |
| 8 | GPT-4o | 44.12 | #343 |
| 9 | Llama 2 70B Chat | 44.03 | #1237 |
| 10 | Mistral 7B Instruct | 42.71 | #1363 |
| 11 | GPT-OSS-20B | 39.96 | #518 |
| 12 | GPT-3.5 Turbo | 39.8 | #865 |
| 13 | GPT-4 | 35.5 | #438 |
| 14 | Claude 2.1 | 34.09 | #839 |
| 15 | Gemma 1.1 7B (IT) | 18.65 | #1368 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=fanoutqa · How It Works · Data refreshed daily, snapshot 2026-10-20.