TOC-Bench - Duration Category: leaderboard

The duration category questions of TOC-Bench (2026), video questions grounded in per-frame object tracks and kept only when not answerable from text alone, single frames or shuffled frames: how long an object event lasts, from very briefly to an extended period. Accuracy in percent; higher is better.

Metric: Duration Category accuracy (%) on TOC-Bench (2,323 human-verified questions over 1,951 videos from Charades, Perception Test, OVIS and MOSE, grounded in per-frame object tracks and kept only when not answerable from text alone, single frames or shuffled frames), mostly 32 uniformly sampled frames, deterministic scoring of multiple-choice, statement-pair, numeric and ordering formats; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.544.6#54
2Seed 2.0 Lite43.9#300
3GLM-5V Turbo42.3#360
4Gemini 3.1 Pro (Preview)40.1#99
5Grok 4.339.7#342
6Kimi K2.630.1#192
7Gemini 3.1 Flash Lite (Preview)30.1#417
8Qwen 2.5 VL 72B30.1#781
9MiMo-V2-Omni26.9#465
10Qwen 3 VL 8B (Thinking)24.7#910
11Video-LLaVA-7B23.7#3879
12GPT-5.4 Mini21.8#406
13LLaVA-Video-7B21.5#2276
14LLaVA-Video-72B21.5#1193
15InternVL3-8B20.2#1782

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=toc-bench-duration-category · How It Works · Data refreshed daily, snapshot 2026-11-21.