Mathematical Reasoning on Math Vision (Pass@1)
64.1Pass@1Base
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen3-VL-Thinking, Model Size=32B2026.05 | 64.1 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 60.6 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 60.4 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 58.9 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 58.4 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 57.9 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 57.4 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 57.2 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=32B2026.05 | 56.5 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 56.1 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 55.3 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 55.3 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 54.9 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 54.4 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 52.8 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 52.5 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 51.6 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 50.9 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 50.8 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 49.1 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=4B2026.05 | 48 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=8B2026.05 | 47.1 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 38.9 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 35.8 | |
| SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 35 | |
| SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 34.8 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 34.6 | |
| GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 33.7 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 30.2 | |
| SFTBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 28.9 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 18.6 | |
| BaseBackbone=Qwen3-VL-Thinking, Model Size=2B2026.05 | 16.9 |