EgoCoT-Bench - Hand-Object Grounded CoT: leaderboard

EgoCoT-Bench (2026) Hand-Object Grounded CoT subtask: from an egocentric manipulation video, answer through a reasoning chain grounded in hand-object and temporal cues. 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 Grounded CoT subtask (answer through a reasoning chain grounded in hand-object and temporal cues); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.284.03#190
2Qwen 3.5 397B A17B83.61#270
3Qwen 3.5 Plus82.77#250
4Qwen 3.5 27B80.67#380
5GPT-5.179.83#248
6Qwen 3 VL 32B79.83#624
7Qwen 3.5 122B A10B79.41#350
8Qwen 3 VL 235B A22B78.99#486
9Qwen 3 VL 8B78.15#927
10Qwen 3 VL Plus77.54#514
11InternVL3.5-4B75.63#1946
12InternVL3.5-8B73.11#1618
13Qwen 3 VL 30B A3B73.11#791
14LLaVA-OneVision-1.5-4B73.11#1874
15InternVL3.5-14B72.27#1519

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-grounded-cot · How It Works · Data refreshed daily, snapshot 2026-11-22.