RefereeBench: leaderboard
RefereeBench (2026) tests multimodal models as sports referees on 925 officiating videos across 11 sports with 6,475 human-annotated multiple-choice questions. This board is overall accuracy in percent, averaged across sports over the seven question types; higher is better.
Metric: Overall accuracy (%) of multiple-choice sports-refereeing questions over 925 officiating videos in 11 sports (6,475 QA pairs), averaged across sports, full video at 720p; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 1.8 | 61.1 | #252 |
| 2 | Gemini 3 Flash | 60.6 | #175 |
| 3 | Gemini 3 Pro | 60 | #145 |
| 4 | GPT-5 | 54.8 | #178 |
| 5 | Claude Sonnet 4.5 | 47.6 | #261 |
| 6 | GPT-4o | 47.1 | #751 |
| 7 | Qwen 3 VL 8B | 47.1 | #917 |
| 8 | Claude Haiku 4.5 | 45.6 | #583 |
| 9 | InternVL3.5-8B | 41.8 | #1604 |
| 10 | VideoLLaMA3-7B | 36 | #2050 |
| 11 | LLaVA-Video-7B | 27.4 | #2276 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=refereebench · How It Works · Data refreshed daily, snapshot 2026-11-21.