EgoCoT-Bench - Goal-Oriented Object Tracking: leaderboard

EgoCoT-Bench (2026) Goal-Oriented Object Tracking subtask: from an egocentric manipulation video, track an object by its functional role in the manipulation goal. 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), Goal-Oriented Object Tracking subtask (track an object by its functional role in the manipulation goal); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.145.28#248
2GPT-5.242.92#190
3Qwen 3.5 397B A17B38.68#270
4InternVL3.5-4B38.21#1946
5InternVL3.5-14B36.79#1519
6Qwen 3.5 Plus35.85#250
7Qwen 3.5 27B34.43#380
8Qwen 3 VL Plus32.7#514
9InternVL3.5-1B32.55#6140
10Qwen 3.5 122B A10B30.19#350
11Qwen 3 VL 32B27.83#624
12Qwen 3 VL 235B A22B27.36#486
13InternVL3.5-2B26.42#3313
14InternVL3.5-8B25.94#1618
15Qwen 3 VL 8B25#927

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-goal-oriented-object-tracking · How It Works · Data refreshed daily, snapshot 2026-11-22.