AgentVQA Egocentric Videos
Multimodal · 2026-10-22
The 200 AgentVQA questions in the Egocentric Videos domain, covering spatial, perception, environment and planning questions about first-person and indoor-scene videos, including the next step toward a task goal. Questions are manually verified four-option multiple-choice items converted from 23 existing datasets, so offline answers stand in for online agent actions such as choosing the next GUI click or predicting a robot's next step. Models answer with a single letter from the visual input (images, screenshots or video) and any action history; chance is 25%. The score is the percentage of these questions answered correctly.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.7 Flash | 72.5 |
| GPT-5.6 Sol | 69.0 |
| Qwen3.8 27B | 69.0 |
| Gemini 3.7 Flash | 67.0 |
| GLM-4.5V | 65.0 |
| GPT-5.6 Luna | 65.0 |
| Qwen3.5 397B A17B | 65.0 |
| Gemini 3.1 Pro Preview | 64.0 |
| Kimi K3 | 64.0 |
| Qwen3.5 35B A3B | 62.0 |
| Opus 4.8 | 60.0 |
| Qwen3.5 122B A10B | 60.0 |
| Qwen3.7-Plus (source unspecified) | 60.0 |
| Qwen3.5-27B | 57.0 |
| GLM-4.6V | 54.0 |
| GLM-4.1V-9B-Thinking | 53.0 |
| Llama 4 Maverick Instruct | 49.0 |
| Sonnet 5 | 47.0 |
| GPT-4o (source unspecified) | 40.0 |