Severity level classification on PANDABENCH (Hard set)
33AccuracyPANDA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PANDA2026.04 | 33 | 34 | 33 | 33 | |
| DepictQA†Access=Open-source, Training=Trained on PANDASET2026.04 | 30 | 24 | 23 | 22 | |
| GPT-5 MiniAccess=Closed-source2026.04 | 27 | 22 | 21 | 20 | |
| SeagullAccess=Open-source2026.04 | 25 | 25 | 21 | 18 | |
| Gemma-3 27BAccess=Open-source2026.04 | 25 | 26 | 25 | 22 | |
| GPT-5 NanoAccess=Closed-source2026.04 | 25 | 20 | 18 | 17 | |
| RandomType=Baseline2026.04 | 25 | 25 | 25 | 24 | |
| Q-InsightAccess=Open-source2026.04 | 24 | 16 | 21 | 12 | |
| DepictQAAccess=Open-source2026.04 | 24 | 17 | 20 | 10 | |
| Linear ProbeType=Baseline2026.04 | 24 | 29 | 26 | 14 | |
| Attentive ProbeType=Baseline2026.04 | 24 | 25 | 25 | 23 | |
| GPT-4oAccess=Closed-source2026.04 | 23 | 21 | 18 | 18 | |
| Gemini 2.5 ProAccess=Closed-source2026.04 | 17 | 21 | 13 | 16 |