Painting Grading on HanMo-Bench
0.631MAEHanMoVLM (RFT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HanMoVLM (RFT)Size=8B2026.03 | 0.631 | 0.8912 | 46.67 | 0.5598 | |
| HanMoVLM (SFT)Size=8B2026.03 | 0.645 | 0.9027 | 43.67 | 0.5531 | |
| HanMoVLM (RFT)Size=4B2026.03 | 0.6632 | 0.9002 | 42.5 | 0.5568 | |
| HanMoVLM (SFT)Size=4B2026.03 | 0.685 | 0.921 | 39.17 | 0.5506 | |
| Gemini-3-ProSize=-, Expert CoT=true2026.03 | 0.9416 | 1.3297 | 36.83 | 0.3101 | |
| Gemini-3-FlashSize=-, Expert CoT=true2026.03 | 0.9466 | 1.3351 | 37 | 0.3035 | |
| Gemini-3-FlashSize=-, Thinking=true2026.03 | 0.9633 | 1.4355 | 34.83 | 0.3052 | |
| Gemini-3-ProSize=-, Thinking=true2026.03 | 0.9783 | 1.3311 | 35.17 | 0.2855 | |
| Gemini-3-ProSize=-2026.03 | 0.9866 | 1.3589 | 33.83 | — | |
| Gemini-3-FlashSize=-2026.03 | 0.9883 | 1.3472 | 31.17 | — | |
| Qwen3-VL-InstructSize=8B, Thinking=true2026.03 | 1.505 | 1.937 | 19 | 0.3402 | |
| Qwen3-VL-InstructSize=8B, Expert CoT=true2026.03 | 1.5124 | 1.9476 | 21.5 | 0.3757 | |
| Qwen3-VL-InstructSize=8B2026.03 | 1.5582 | 1.9534 | 18.53 | — | |
| GPT-4Size=-, Expert CoT=true2026.03 | 1.7491 | 2.1683 | 17.41 | 0.3049 | |
| GPT-5.1Size=-, Expert CoT=true2026.03 | 1.7671 | 2.1471 | 18.41 | 0.3073 | |
| GPT-5.1Size=-2026.03 | 1.7742 | 2.2123 | 17.73 | — | |
| GPT-4Size=-2026.03 | 1.8023 | 2.2387 | 17.06 | — | |
| GPT-5.1Size=-, Thinking=true2026.03 | 1.8204 | 2.52 | 17.95 | 0.1648 | |
| InternVL-3.5Size=8B2026.03 | 1.8866 | 2.3579 | 17.83 | 0.3395 | |
| Qwen3-VL-InstructSize=4B, Expert CoT=true2026.03 | 1.9341 | 2.4495 | 19.44 | 0.2919 | |
| Qwen3-VL-InstructSize=4B, Thinking=true2026.03 | 1.9493 | 2.4465 | 17.89 | 0.2836 | |
| InternVL-3.5Size=4B2026.03 | 1.9591 | 2.4382 | 13.83 | 0.3441 | |
| Qwen3-VL-InstructSize=4B2026.03 | 1.9734 | 2.4055 | 7.9 | — | |
| GLM4.6V-FlashSize=9B2026.03 | 2.0495 | 2.5466 | 15.5 | 0.3517 | |
| STEP3-VLSize=10B2026.03 | 2.295 | 2.7823 | 15.67 | 0.2335 |