Perception on V* (Pass@1)
90.2Pass@1SFT + GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 90.2 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=32B2026.05 | 89.1 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 88.9 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 87.8 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 87.7 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 87 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 85.7 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 84.8 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 82.7 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 81.7 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 81.3 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 80.6 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 77.7 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 75.9 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 67.1 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 63 |