TAR-Bench - Multiple-Choice QA: leaderboard

Metric: Accuracy (%; multiple-choice questions about the anomaly; 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 December 2026. 11 models tracked.

Top models

#ModelScore
1Qwen 3.5 27B (Non-reasoning)85
2Qwen 3.5 9B (Non-reasoning)82.5
3Gemini 3.1 Pro (Preview)80.4
4Gemma 4 31B (IT)77.5
5Qwen 3 VL 32B Instruct71.2
6Qwen 3 VL 8B Instruct62.9

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