Reasoning on Math Vision (Pass@4)
75.7Pass@4AXPO
Evaluation results
| Method | Links | |
|---|---|---|
| AXPOModel Size=8B, Backbone=Qwen3-VL-Thinking2026.05 | 75.7 | |
| SFT + GRPOModel Size=8B, Backbone=Qwen3-VL-Thinking2026.05 | 75 | |
| AXPOModel Size=4B, Backbone=Qwen3-VL-Thinking2026.05 | 74.3 | |
| BaseModel Size=32B, Backbone=Qwen3-VL-Thinking2026.05 | 73 | |
| SFTModel Size=8B, Backbone=Qwen3-VL-Thinking2026.05 | 71.4 | |
| SFT + GRPOModel Size=4B, Backbone=Qwen3-VL-Thinking2026.05 | 69.7 | |
| GRPOModel Size=8B, Backbone=Qwen3-VL-Thinking2026.05 | 69.7 | |
| SFTModel Size=4B, Backbone=Qwen3-VL-Thinking2026.05 | 69.4 | |
| GRPOModel Size=4B, Backbone=Qwen3-VL-Thinking2026.05 | 69.1 | |
| BaseModel Size=4B, Backbone=Qwen3-VL-Thinking2026.05 | 64.8 | |
| BaseModel Size=8B, Backbone=Qwen3-VL-Thinking2026.05 | 63.5 | |
| AXPOModel Size=2B, Backbone=Qwen3-VL-Thinking2026.05 | 61.2 | |
| SFT + GRPOModel Size=2B, Backbone=Qwen3-VL-Thinking2026.05 | 55.9 | |
| GRPOModel Size=2B, Backbone=Qwen3-VL-Thinking2026.05 | 53.3 | |
| SFTModel Size=2B, Backbone=Qwen3-VL-Thinking2026.05 | 53 | |
| BaseModel Size=2B, Backbone=Qwen3-VL-Thinking2026.05 | 32.6 |