Document Understanding on DocVQA (val)
95.45AccuracyERNIE 5.0
Evaluation Results
| Method | Links | |
|---|---|---|
| ERNIE 5.02026.02 | 95.45 | |
| Qwen3-VLmode=Thinking2026.02 | 95.44 | |
| GPT-5tier=High2026.02 | 94.16 | |
| Gemini 2.5-Pro2026.02 | 91.43 | |
| Gemini 3-Pro2026.02 | 90.7 | |
| Vanilla RoPEPositional Encoding Method=Vanilla RoPE, DIPE Enhancement=+DIPE2026.03 | 40.18 | |
| MRoPE-IPositional Encoding Method=MRoPE-I, DIPE Enhancement=+DIPE2026.03 | 39.9 | |
| MRoPEPositional Encoding Method=MRoPE, DIPE Enhancement=+DIPE2026.03 | 38.55 | |
| Vanilla RoPEPositional Encoding Method=Vanilla RoPE, DIPE Enhancement=Base2026.03 | 38.38 | |
| MRoPE-IPositional Encoding Method=MRoPE-I, DIPE Enhancement=Base2026.03 | 36.01 | |
| MRoPEPositional Encoding Method=MRoPE, DIPE Enhancement=Base2026.03 | 33.25 |