EgoCoT-Bench - Hand-Object Temporal Retrospection: leaderboard

EgoCoT-Bench (2026) Hand-Object Temporal Retrospection subtask: from an egocentric manipulation video, order earlier hand-object interactions in time. Four-way multiple choice accuracy in percent; higher is better.

Metric: Answer accuracy (%, four-way multiple choice, strict exact match) of zero-shot MLLMs on egocentric manipulation videos of EgoCoT-Bench (351 clips, 3,172 human-verified questions built from spatio-temporal scene graphs), Hand-Object Temporal Retrospection subtask (order earlier hand-object interactions in time); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-14B75#1519
2InternVL3.5-1B75#6140
3InternVL3.5-2B71.88#3313
4InternVL3.5-8B68.75#1618
5Qwen 3 VL 32B68.75#624
6InternVL3.5-4B62.5#1946
7Qwen 3 VL 30B A3B62.5#791
8GPT-5.259.38#190
9Qwen 3.5 27B59.38#380
10Qwen 3 VL 8B59.38#927
11Qwen 3.5 397B A17B56.25#270
12Qwen 3.5 122B A10B56.25#350
13GPT-5.156.25#248
14Qwen 3 VL 235B A22B56.25#486
15Qwen 3 VL Plus56.25#514

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

Interactive version: theaggregate.ai/benchmark?slug=egocot-bench-hand-object-temporal-retrospection · How It Works · Data refreshed daily, snapshot 2026-11-22.