EgoCoT-Bench - Local Cause-Outcome Inference: leaderboard

EgoCoT-Bench (2026) Local Cause-Outcome Inference subtask: from an egocentric manipulation video, identify the recent action that caused an observed outcome. 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), Local Cause-Outcome Inference subtask (identify the recent action that caused an observed outcome); 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 Plus74.72#242
2Qwen 3.5 122B A10B73.61#346
3Qwen 3.5 27B73.33#375
4GPT-5.272.22#185
5Qwen 3 VL Plus71.94#512
6Qwen 3 VL 32B71.67#622
7Qwen 3.5 397B A17B70.83#264
8GPT-5.170.56#240
9InternVL3.5-14B70#1504
10InternVL3.5-8B67.22#1604
11Qwen 3 VL 30B A3B66.94#784
12Qwen 3 VL 8B64.72#917
13InternVL3.5-4B63.06#1930
14Qwen 3 VL 235B A22B62.5#485
15LLaVA-OneVision-1.5-8B61.94#1864

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-local-cause-outcome-inference · How It Works · Data refreshed daily, snapshot 2026-11-21.