Cultural Moment Benchmark - Temporal Localization: leaderboard
Metric: Mean IoU (%; Stage 3, predicted start and end time of a described sub-event on a different example video against the annotated span, averaged over 631 Stage 3 pairs; 306 expert-curated concepts from seven Southeast Asian countries in five categories (celebrations, dances, games, music, weddings); Carry mode, where the model's own previous-stage answer is carried into the next stage; Gemini, GPT and Qwen read video at 1 fps, InternVL at its native 12 frames). Source: arxiv.org. Saturation forecast: Around 2029. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 33.4 |
| 2 | GPT-5.4 | 31 |
| 3 | Qwen 3.5 27B | 24.1 |
| 4 | Qwen 3 VL 32B Instruct | 22.3 |
Interactive version: theaggregate.ai/benchmark?slug=cultural-moment-benchmark-temporal-localization · How It Works · Data refreshed daily, snapshot 2026-09-26.