Image Captioning on DecapBench (test)
87.8PrecisionVCap (e1)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VCap (e1)Iteration=e12026.05 | 87.8 | 58.83 | 69.52 | 73.32 | |
| VCap (e2)Iteration=e22026.05 | 86.42 | 60.92 | 70.35 | 73.67 | |
| GPT-5.42026.05 | 86.27 | 60.08 | 70.1 | 73.18 | |
| Gemini 3.1 Pro2026.05 | 83.76 | 56.63 | 66.46 | 70.2 | |
| Qwen3-VL-8B-InstrModel size=8B, Type=Instruction-tuned2026.05 | 82.73 | 55.61 | 65.7 | 69.17 | |
| Qwen3-VL-235B-InstrModel size=235B, Type=Instruction-tuned2026.05 | 79.76 | 55.99 | 64.99 | 67.88 | |
| Qwen3.5-397BModel size=397B2026.05 | 78.15 | 56.21 | 64.53 | 67.18 | |
| Seed 2.0 Pro2026.05 | 77.74 | 56.95 | 64.89 | 67.35 | |
| CapRL2026.05 | 75.93 | 50.09 | 59.47 | 63.01 |