TAR-Bench - Binary QA with Explanation: leaderboard

Metric: Scaled BERTScore F1 (%; a yes/no answer with an explanation, compared with the reference; 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 2033. 11 models tracked.

Top models

#ModelScore
1Qwen 3.5 9B (Non-reasoning)50.7
2Gemma 4 31B (IT)48
3Qwen 3.5 27B (Non-reasoning)45.3
4Gemini 3.1 Pro (Preview)44.1
5Qwen 3 VL 32B Instruct42.7
6Qwen 3 VL 8B Instruct39

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