Distortion Identification on PANDABENCH Easy
78AccuracyPANDA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PANDAModel Type=Open-source2026.04 | 78 | 79 | 81 | 79 | |
| DepictQAModel Type=Open-source, Trained on PANDASET=true2026.04 | 75 | 82 | 71 | 76 | |
| GPT-5 MiniModel Type=Closed-source2026.04 | 49 | 54 | 44 | 44 | |
| GPT-4oModel Type=Closed-source2026.04 | 46 | 60 | 41 | 44 | |
| Attentive ProbeModel Type=Baseline2026.04 | 40 | 38 | 42 | 39 | |
| Gemini 2.5 ProModel Type=Closed-source2026.04 | 39 | 59 | 36 | 41 | |
| GPT-5 NanoModel Type=Closed-source2026.04 | 37 | 37 | 29 | 28 | |
| Gemma-3Model Type=Open-source, Parameters=27B2026.04 | 27 | 31 | 24 | 20 | |
| SeagullModel Type=Open-source2026.04 | 23 | 25 | 20 | 18 | |
| Linear ProbeModel Type=Baseline2026.04 | 20 | 16 | 9 | 7 | |
| Q-SiTModel Type=Open-source2026.04 | 18 | 9 | 7 | 5 | |
| Q-InsightModel Type=Open-source2026.04 | 16 | 16 | 9 | 7 | |
| DepictQAModel Type=Open-source2026.04 | 15 | 11 | 13 | 10 | |
| RandomModel Type=Baseline2026.04 | 7 | 7 | 7 | 6 |