EgoCoT-Bench - High-level Grounded Reasoning: leaderboard

EgoCoT-Bench (2026) High-level Grounded Reasoning group (progress reasoning, hand-object grounded chain of thought and goal-oriented object tracking) 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), High-level Grounded Reasoning group (progress reasoning, hand-object grounded chain of thought and goal-oriented object tracking); 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 397B A17B68.08#264
2Qwen 3.5 Plus67.64#242
3Qwen 3.5 27B65.3#375
4GPT-5.165.15#240
5GPT-5.264.86#185
6Qwen 3.5 122B A10B64.28#346
7Qwen 3 VL Plus60.53#512
8InternVL3.5-4B60.32#1930
9Qwen 3 VL 235B A22B60.18#485
10Qwen 3 VL 32B60.03#622
11InternVL3.5-14B57.54#1504
12InternVL3.5-8B56.37#1604
13Qwen 3 VL 8B55.43#917
14LLaVA-OneVision-1.5-8B54.76#1864
15LLaVA-OneVision-1.5-4B53.88#1855

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-high-level-grounded-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.