TOC-Bench - Reappearance Identity: leaderboard

The reappearance identity 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: whether an object that reappears is the same instance. Accuracy in percent; higher is better.

Metric: Reappearance Identity 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: Not forecast. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1MiniCPM-V-2.6100#3311
2Qwen 3 VL 8B (Thinking)100#924
3LLaVA-Video-7B100#2303
4Video-R1100#1384
5VideoChat-R1.5-7B100#1260
6Qwen 2.5 VL 7B95.7#1878
7Grok 4.395.7#351
8Seed 2.0 Lite95.7#306
9Qwen 2.5 VL 72B95.7#782
10GPT-5.591.3#56
11InternVL3-8B91.3#1795
12Gemini 3.1 Flash Lite (Preview)91.3#419
13Gemini 3.1 Pro (Preview)87#102
14Kimi K2.687#197
15VideoLLaMA3-7B78.3#2064

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=toc-bench-reappearance-identity · How It Works · Data refreshed daily, snapshot 2026-11-22.