RobotEQ-Video - Synthetic Videos: leaderboard

Metric: Synthetic-subset Avg (%; 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; 1,218 generated videos, 11,225 actions). Source: arxiv.org. Saturation forecast: Around May 2027. 48 models tracked.

Top models

#ModelScore
1GPT-5.579.8
2Gemini 3.5 Flash79.7
3Seed 2.0 Lite79.4
4GPT-5.6 Luna79.1
5Gemini 3.1 Flash Lite77.8
6Qwen 3 VL 30B A3B Instruct76.7
7Qwen 3.5 27B (Thinking)76.7
8Qwen 3.6 Plus76.2
9Gemma 4 12B (IT)76.1
10Gemma 4 31B (IT)76
11Kimi K2.675
12Claude Sonnet 575
13GLM-5V Turbo75
14Qwen 3.5 9B (Thinking)74.9
15Qwen 3 VL 30B A3B (Thinking)74.5

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