EgoCoT-Bench - Manipulation State Perception: leaderboard

EgoCoT-Bench (2026) Manipulation State Perception subtask: from an egocentric manipulation video, recognize the current manipulation state of the 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), Manipulation State Perception subtask (recognize the current manipulation state of the 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 397B A17B87.79#270
2Qwen 3.5 122B A10B86.38#350
3Qwen 3.5 Plus85.92#250
4Qwen 3.5 27B84.51#380
5Qwen 3 VL 30B A3B82.16#791
6Qwen 3 VL 8B81.6#927
7Qwen 3 VL 32B79.81#624
8InternVL3.5-2B79.81#3313
9GPT-5.177#248
10Qwen 3 VL 235B A22B77#486
11Qwen 3 VL Plus77#514
12InternVL3.5-14B75.12#1519
13LLaVA-OneVision-1.5-8B74.65#1876
14InternVL3.5-4B73.24#1946
15GPT-5.272.3#190

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-manipulation-state-perception · How It Works · Data refreshed daily, snapshot 2026-11-22.