Embodied Question Answering on OpenEQA EM-EQA Episodes up to 32 frames
86.8LLM-Match ScoreHuman
Evaluation Results
| Method | Links | |
|---|---|---|
| Human2025.10 | 86.8 | |
| Gemini 3 FlashQuery Latency [sec]=10.52025.10 | 76.8 | |
| Qwen 3.5 Plus2025.10 | 74.1 | |
| Gemini 2.5 FlashQuery Latency [sec]=6.82025.10 | 69.8 | |
| VL-KnG (GER-L)Query Latency [sec]=0.82025.10 | 55.2 | |
| VL-KnG (GER-G)Query Latency [sec]=0.82025.10 | 54.7 | |
| VL-KnG (GR)Query Latency [sec]=0.82025.10 | 50.7 | |
| GPT-4V2025.10 | 49.6 | |
| Gemini 1.0 Pro Vision2025.10 | 44.9 | |
| GPT-4 w/ ConceptGraphs2025.10 | 36.5 |