RobotEQ-Video: leaderboard

Metric: A-Avg: mean of the synthetic-subset and real-subset averages (%; action-properness judgment: given an egocentric robot-view video and a numbered list of candidate robot actions, the model labels each action proper or improper for socially appropriate proactive assistance; score is the mean of accuracy and macro-F1 against the majority-voted human label and of probability of agreement and consensus-weighted agreement with the four individual annotators). Source: arxiv.org. Saturation forecast: Around December 2027. 48 models tracked.

Top models

#ModelScore
1GPT-5.6 Luna79.5
2GPT-5.579.3
3Gemini 3.5 Flash78.9
4Seed 2.0 Lite78.8
5Gemini 3.1 Flash Lite77.4
6Qwen 3.5 27B (Thinking)77.1
7Qwen 3 VL 30B A3B Instruct76
8Gemma 4 12B (IT)75.7
9Gemma 4 31B (IT)75.2
10Kimi K2.675.2
11Qwen 3.6 Plus74.9
12Qwen 3.5 9B (Thinking)74.5
13Qwen 3 VL 30B A3B (Thinking)74.4
14Qwen 3 VL 8B (Thinking)74.3
15GLM-4.1V-9B (Thinking)74.1

Interactive version: theaggregate.ai/benchmark?slug=roboteq-video · How It Works · Data refreshed daily, snapshot 2026-09-26.