Vision-Language Reasoning on VL Reasoning Suite (MathVista, MMMU, MMStar)
79.8MathVista AccFRISM
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FRISMBase Model=Qwen2.5-VL-32B-Instruct, Reasoning Model=QwQ-32B2026.01 | 79.8 | 49.3 | 42.6 | 67 | 51.9 | 67.8 | 59.7 | |
| TABase Model=Qwen2.5-VL-32B-Instruct, Reasoning Model=QwQ-32B2026.01 | 78.1 | 48.2 | 40.3 | 67.6 | 50 | 67.7 | 58.7 | |
| BaseBase Model=Qwen2.5-VL-32B-Instruct, Reasoning Model=QwQ-32B2026.01 | 77.7 | 47 | 38.2 | 65.2 | 47.5 | 67.9 | 57.3 | |
| IP-MergingBase Model=Qwen2.5-VL-32B-Instruct, Reasoning Model=QwQ-32B2026.01 | 75.5 | 45.2 | 38.4 | 65.9 | 48.3 | 67.9 | 56.9 | |
| Ties-MergingBase Model=Qwen2.5-VL-32B-Instruct, Reasoning Model=QwQ-32B2026.01 | 74.8 | 42.8 | 35.2 | 63.9 | 42.1 | 65.3 | 54 | |
| FRISMBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B2026.01 | 70.3 | 41.8 | 26.4 | 57.8 | 37.5 | 62.4 | 49.4 | |
| IP-MergingBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Temperature (T)=0.32026.01 | 68.8 | 40.6 | 25 | 54.7 | 35.9 | 60.1 | 47.5 | |
| DAREBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B2026.01 | 68.4 | 41.1 | 25.4 | 56.4 | 34.6 | 60.8 | 47.8 | |
| BaseBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B2026.01 | 68.1 | 41.2 | 25.6 | 55.1 | 34.6 | 60.1 | 47.4 | |
| IP-MergingBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Temperature (T)=0.42026.01 | 67.2 | 40.8 | 25.5 | 55.9 | 36 | 60.7 | 47.7 | |
| IP-MergingBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Temperature (T)=0.22026.01 | 65.9 | 38.7 | 24.9 | 55.1 | 33.9 | 58.7 | 46.2 | |
| TABase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Lambda (λ)=0.12026.01 | 65.4 | 37.8 | 24.2 | 53.7 | 36.6 | 60.7 | 46.4 | |
| Ties-MergingBase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B2026.01 | 61.9 | 38.8 | 23.5 | 53 | 35.7 | 58.6 | 45.3 | |
| FRISMBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B2026.01 | 45 | 28.3 | 21.4 | 42.8 | 28.9 | 43.4 | 35 | |
| Ties-MergingBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B2026.01 | 41.8 | 17.8 | 17.4 | 45 | 25.9 | 41.4 | 31.6 | |
| TABase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Lambda (λ)=0.12026.01 | 41.5 | 26.7 | 20.6 | 39 | 28.8 | 40.7 | 32.9 | |
| TABase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Lambda (λ)=0.22026.01 | 41.4 | 28 | 21.2 | 37.8 | 27.5 | 42.1 | 33 | |
| IP-MergingBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Temperature (T)=0.32026.01 | 41.1 | 24.1 | 18.5 | 42 | 27 | 40.7 | 32.2 | |
| IP-MergingBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Temperature (T)=0.42026.01 | 41.1 | 24.1 | 18.5 | 42 | 27 | 40.7 | 32.2 | |
| DAREBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B2026.01 | 40.7 | 27.1 | 20.5 | 41 | 25.8 | 40.4 | 32.6 | |
| BaseBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B2026.01 | 40.5 | 25.5 | 20.6 | 44 | 28 | 40.5 | 33.2 | |
| TABase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Lambda (λ)=0.152026.01 | 38.5 | 27.9 | 19.7 | 39.2 | 29 | 43.2 | 32.9 | |
| TABase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Lambda (λ)=0.152026.01 | 32.9 | 17.1 | 10.7 | 38.3 | 22 | 40.3 | 26.9 | |
| TABase Model=Qwen2.5-VL-7B-Instruct, Reasoning Model=DeepSeek-R1-Distill-Qwen-7B, Lambda (λ)=0.22026.01 | 28.2 | 14.8 | 9.2 | 34.6 | 21.2 | 34.6 | 23.8 | |
| IP-MergingBase Model=Qwen2.5-VL-3B-Instruct, Reasoning Model=SmallThinker-3B, Temperature (T)=0.22026.01 | 21.8 | 15.8 | 8.6 | 25.6 | 14.1 | 27 | 18.8 |