Rationale Alignment on WiserUI-Bench
69.05Interp. RecallPERCEPTUI
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PERCEPTUI2026.06 | 69.05 | 84.09 | |
| PERCEPTUIablation=w/o RPE2026.06 | 66.42 | 80.11 | |
| o12026.06 | 64.18 | 78.33 | |
| GPT-52026.06 | 53.78 | 69.33 | |
| Claude Opus 4.62026.06 | 52.11 | 67.99 | |
| Qwen2.5-VLParameters=32B2026.06 | 50.23 | 63 | |
| GPT-4o2026.06 | 50.15 | 66.67 | |
| PERCEPTUIablation=w/o CR2026.06 | 47.86 | 68.22 | |
| InternVL-2.5Parameters=38B2026.06 | 45.32 | 59.33 | |
| Qwen2.5-VLParameters=7B2026.06 | 43.71 | 61.67 | |
| InternVL-2.5Parameters=8B2026.06 | 35.09 | 51.67 | |
| LLaVA-OneVisionParameters=7B2026.06 | 21.78 | 33.33 | |
| LLaVA-NeXTParameters=7B2026.06 | 11.99 | 21.67 |