EgoCoT-Bench - Progress Reasoning: leaderboard

EgoCoT-Bench (2026) Progress Reasoning subtask: from an egocentric manipulation video, infer the current operation step or whether a step is complete. 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), Progress Reasoning subtask (infer the current operation step or whether a step is complete); 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 Plus81.12#242
2Qwen 3.5 122B A10B79.83#346
3Qwen 3.5 397B A17B78.97#264
4Qwen 3.5 27B77.68#375
5Qwen 3 VL 235B A22B70.82#485
6Qwen 3 VL 32B69.1#622
7LLaVA-OneVision-1.5-8B68.67#1864
8Qwen 3 VL Plus68.53#512
9GPT-5.168.24#240
10InternVL3.5-8B66.95#1604
11Qwen 3 VL 30B A3B66.52#784
12GPT-5.265.24#185
13InternVL3.5-4B64.81#1930
14LLaVA-OneVision-1.5-4B63.95#1855
15InternVL3.5-14B61.37#1504

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-progress-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.