EgoCoT-Bench - Spatio-Temporal Retrospection: leaderboard

EgoCoT-Bench (2026) Spatio-Temporal Retrospection group (state, location or visibility, and hand-object temporal retrospection) on egocentric manipulation videos. 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), Spatio-Temporal Retrospection group (state, location or visibility, and hand-object temporal retrospection); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.5 27B69.1#375
2Qwen 3 VL 30B A3B65.49#784
3Qwen 3.5 122B A10B65.11#346
4InternVL3.5-2B64.18#3284
5InternVL3.5-8B63.3#1604
6Qwen 3.5 Plus62.67#242
7Qwen 3.5 397B A17B62.55#264
8GPT-5.262.42#185
9InternVL3.5-14B61.92#1504
10Qwen 3 VL 8B61.75#917
11Qwen 3 VL Plus59.75#512
12Qwen 3 VL 235B A22B59.14#485
13Qwen 3 VL 32B58.76#622
14LLaVA-OneVision-1.5-4B56.62#1855
15GPT-5.155.23#240

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

Interactive version: theaggregate.ai/benchmark?slug=egocot-bench-spatio-temporal-retrospection · How It Works · Data refreshed daily, snapshot 2026-11-21.