In-the-wild model generalization on Human Bench Vision-based Demo
15.5NSEProgressLM-RL-3B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ProgressLM-RL-3BModel Scale=3B, Training Protocol=RL, Batch Size=82026.01 | 15.5 | 88.9 | 90 | |
| Qwen3VL-8BModel Scale=8B, Batch Size=82026.01 | 19.6 | 80.9 | 0 | |
| Qwen3VL-32BModel Scale=32B, Batch Size=82026.01 | 19.7 | 91.7 | 0 | |
| ProgressLM-SFT-3BModel Scale=3B, Training Protocol=SFT, Batch Size=82026.01 | 19.7 | 76.4 | 320 | |
| Qwen3VL-4BModel Scale=4B, Batch Size=82026.01 | 22.4 | 80.4 | 0 | |
| Qwen2.5VL-72BModel Scale=72B, Batch Size=82026.01 | 23.5 | 78.5 | 20 | |
| Qwen2.5VL-3BModel Scale=3B, Batch Size=82026.01 | 27.8 | 48 | 0 | |
| Qwen2.5VL-7BModel Scale=7B, Batch Size=82026.01 | 29 | 52.2 | 60 | |
| Qwen2.5VL-32BModel Scale=32B, Batch Size=82026.01 | 31.7 | 50 | 0 | |
| Intern3.5-VL-14BModel Scale=14B, Batch Size=82026.01 | 34.4 | 52.2 | 800 | |
| Intern3.5-VL-4BModel Scale=4B, Batch Size=82026.01 | 34.7 | 50.3 | 0 | |
| Intern3.5-VL-38BModel Scale=38B, Batch Size=82026.01 | 40.8 | 44.5 | 1,420 | |
| Qwen3VL-2BModel Scale=2B, Batch Size=82026.01 | 48.4 | 33.6 | 10 | |
| Intern3.5-VL-8BModel Scale=8B, Batch Size=82026.01 | 64.9 | 15.7 | 20 |