EgoCoT-Bench - Location and Visibility Retrospection: leaderboard

EgoCoT-Bench (2026) Location and Visibility Retrospection subtask: from an egocentric manipulation video, recover an earlier location or visibility of an object. 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), Location and Visibility Retrospection subtask (recover an earlier location or visibility of an object); 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 27B67.74#375
2InternVL3.5-8B67.74#1604
3InternVL3.5-2B66.94#3284
4Qwen 3 VL 30B A3B65.86#784
5InternVL3.5-14B63.71#1504
6Qwen 3.5 122B A10B62.7#346
7Qwen 3.5 Plus60.69#242
8Qwen 3 VL 8B60.69#917
9GPT-5.259.88#185
10Qwen 3.5 397B A17B59.07#264
11Qwen 3 VL 32B55.04#622
12LLaVA-OneVision-1.5-4B55.04#1855
13Qwen 3 VL Plus54.66#512
14InternVL3.5-4B54.44#1930
15Qwen 3 VL 235B A22B52.82#485

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-location-and-visibility-retrospection · How It Works · Data refreshed daily, snapshot 2026-11-21.