Perception on HRBen 8K
78.9Pass@1Base
Evaluation results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen3-VL-Thinking, Model Size=32B2026.05 | 78.9 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 78.3 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 77 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 74.9 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 74.4 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 74.1 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 73.9 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 73.8 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 72.4 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 71.1 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 70.6 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 68.8 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 67.5 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 66.1 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 59.4 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 55.9 |