EgoCoT-Bench - Predictive and Causal Inference: leaderboard

EgoCoT-Bench (2026) Predictive and Causal Inference group (next state and next action anticipation and local cause-outcome inference) 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), Predictive and Causal Inference group (next state and next action anticipation and local cause-outcome inference); 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 Plus78.21#242
2Qwen 3.5 27B77.03#375
3GPT-5.176.63#240
4Qwen 3 VL 32B76.53#622
5Qwen 3.5 122B A10B76.32#346
6Qwen 3.5 397B A17B76.11#264
7Qwen 3 VL Plus76#512
8GPT-5.275.68#185
9Qwen 3 VL 235B A22B73.37#485
10InternVL3.5-14B72.53#1504
11InternVL3.5-8B72.42#1604
12Qwen 3 VL 8B71.91#917
13LLaVA-OneVision-1.5-8B71.79#1864
14Qwen 3 VL 30B A3B71.47#784
15InternVL3.5-4B70#1930

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-predictive-and-causal-inference · How It Works · Data refreshed daily, snapshot 2026-11-21.