TAR-Bench: leaderboard

Metric: Mean score (%; unweighted mean of the ten task scores: accuracy for binary and multiple-choice QA, mean IoU for temporal localization, scaled BERTScore F1 for the seven open-ended tasks; zero-shot in non-reasoning (direct-answer) mode through VLMEvalKit, mean of 3 runs, on TAR-Bench: 960 human-curated annotations for 80 held-out traffic-anomaly clips from 17 public YouTube videos). Source: arxiv.org. Saturation forecast: Around 2034. 11 models tracked.

Top models

#ModelScore
1Qwen 3.5 27B (Non-reasoning)39.5
2Qwen 3.5 9B (Non-reasoning)39.1
3Gemini 3.1 Pro (Preview)36.3
4Gemma 4 31B (IT)33.3
5Qwen 3 VL 8B Instruct30.9
6Qwen 3 VL 32B Instruct28

Interactive version: theaggregate.ai/benchmark?slug=tar-bench · How It Works · Data refreshed daily, snapshot 2026-09-29.