SPRINT - Cause Identification: leaderboard
SPRINT (Renmin University of China, 2026) evaluates whether multimodal LLMs proactively infer physical hazards in real-world sports videos, judged by an automatic evaluator over the full video under an explicit safety-inquiry prompt. Cause identification is the share of videos in which the model states the correct direct physical cause, conditional on factor coverage, in percent, higher is better.
Metric: Cause identification rate (%; the model states the correct direct physical cause, conditional on factor coverage; over SPRINT's 2,440 real-world sports-accident videos (full video); binary per-video credit judged by a Gemini 3 Flash evaluator; explicit safety-inquiry prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 65 | #159 |
| 2 | GPT-5 | 64 | #163 |
| 3 | Seed 1.8 | 64 | #239 |
| 4 | Gemini 3 Pro | 61 | #135 |
| 5 | GPT-4o | 46 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sprint-cause-identification · How It Works · Data refreshed daily, snapshot 2026-11-12.