TOC-Bench - Cross-Object Order: leaderboard

The cross-object order 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: which of two events on different objects happens first. Accuracy in percent; higher is better.

Metric: Cross-Object Order 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 2028. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Kimi K2.663.6#192
2GPT-5.563.1#54
3Grok 4.361#342
4Gemini 3.1 Pro (Preview)57.2#99
5GLM-5V Turbo56.1#360
6Seed 2.0 Lite52.9#300
7MiniCPM-V-2.652.4#3267
8Qwen 3 VL 8B (Thinking)50.8#910
9Qwen 2.5 VL 72B50.3#781
10GPT-5.4 Mini49.2#406
11LLaVA-Video-7B49.2#2276
12Gemini 3.1 Flash Lite (Preview)48.1#417
13MiMo-V2-Omni47.6#465
14Video-LLaVA-7B47.1#3879
15VideoChat-R1.5-7B47.1#1254

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-cross-object-order · How It Works · Data refreshed daily, snapshot 2026-11-21.