Quality scoring on PANDABENCH (Hard set)
0.36SRCCPANDA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PANDA2026.04 | 0.36 | 0.38 | |
| Linear ProbeType=Baseline2026.04 | 0.22 | 0.23 | |
| DepictQA†Access=Open-source, Training=Trained on PANDASET2026.04 | 0.18 | 0.17 | |
| Gemini 2.5 ProAccess=Closed-source2026.04 | 0.1 | 0.14 | |
| GPT-5 MiniAccess=Closed-source2026.04 | 0.09 | 0.13 | |
| GPT-4oAccess=Closed-source2026.04 | 0.06 | 0.08 | |
| GPT-5 NanoAccess=Closed-source2026.04 | 0.02 | 0.04 | |
| Attentive ProbeType=Baseline2026.04 | 0.02 | 0.02 | |
| RandomType=Baseline2026.04 | 0 | 0 |