Mathematical Reasoning on MathVision (AUC %)
95.68AUC (%)ReLope
Evaluation Results
| Method | Links | |
|---|---|---|
| ReLopeBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 95.68 | |
| ReLopeBackbone=Phi-4-Multimodal-Instruct2026.03 | 91.16 | |
| ProbeBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 90.28 | |
| Attention ProbeBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 88.92 | |
| Post-Hoc EmbedBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 88.14 | |
| RouteLLMBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 87.53 | |
| Attention ProbeBackbone=Phi-4-Multimodal-Instruct2026.03 | 83.47 | |
| MoTBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 82.59 | |
| ReLopeBackbone=Gemma3-12B2026.03 | 81.77 | |
| ProbeBackbone=Phi-4-Multimodal-Instruct2026.03 | 81.03 | |
| RouteLLMBackbone=Phi-4-Multimodal-Instruct2026.03 | 78.55 | |
| Cascade RoutingBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 78.25 | |
| Post-Hoc EmbedBackbone=Phi-4-Multimodal-Instruct2026.03 | 77.12 | |
| Cascade RoutingBackbone=Phi-4-Multimodal-Instruct2026.03 | 76.84 | |
| MoTBackbone=Phi-4-Multimodal-Instruct2026.03 | 73.28 | |
| ProbeBackbone=Gemma3-12B2026.03 | 72.26 | |
| Attention ProbeBackbone=Gemma3-12B2026.03 | 71.19 | |
| RouteLLMBackbone=Gemma3-12B2026.03 | 70.45 | |
| Post-Hoc EmbedBackbone=Gemma3-12B2026.03 | 69.07 | |
| Cascade RoutingBackbone=Gemma3-12B2026.03 | 68.1 | |
| MoTBackbone=Gemma3-12B2026.03 | 65.39 |