RefereeBench - Temporal Grounding: 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 accuracy in percent on questions about when the infraction happens in the video, averaged across sports; higher is better.
Metric: Accuracy (%) on temporal-grounding questions (Q7), averaged across 11 sports; 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 | 50.1 | #252 |
| 2 | Gemini 3 Pro | 48.8 | #145 |
| 3 | Gemini 3 Flash | 46.8 | #175 |
| 4 | Qwen 3 VL 8B | 39 | #917 |
| 5 | GPT-5 | 38.1 | #178 |
| 6 | Claude Haiku 4.5 | 34.5 | #583 |
| 7 | Claude Sonnet 4.5 | 33.3 | #261 |
| 8 | VideoLLaMA3-7B | 30.1 | #2050 |
| 9 | InternVL3.5-8B | 29.6 | #1604 |
| 10 | GPT-4o | 28.6 | #751 |
| 11 | LLaVA-Video-7B | 27.8 | #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-temporal-grounding · How It Works · Data refreshed daily, snapshot 2026-11-21.