Open-ended Question Answering on PBSBench Slide-level 1.0 (test)
36BLEU-1PBS-VL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PBS-VLFramework=Slide-level, Vision backbone=Dinobloom ViT-L/14, VLM backbone=Qwen 2.5 VL-7B2026.04 | 36 | 39 | 76 | |
| GPT-4oInput=30 random patches2026.04 | 17 | 20 | 73 | |
| SlideChatModel type=Pathology-specific for WSIs2026.04 | 16 | 20 | 62 | |
| Claude-4.5Input=30 random patches2026.04 | 15 | 18 | 67 | |
| Gemini-2.5-proInput=30 random patches2026.04 | 9 | 12 | 68 | |
| HistoGPTModel type=Pathology-specific for WSIs2026.04 | 3 | 4 | 15 |