Vision-Language Reasoning on VL Reasoning Benchmarks
74MVista ScoreFRISM
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FRISMMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 74 | 42.8 | 28.1 | 50.9 | 58.7 | 38.4 | 63.1 | |
| FRISMMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 74 | 42.6 | 30.7 | 53.2 | 61 | 41.2 | 69.3 | |
| IP-MergingMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, T=0.42026.01 | 73.4 | 41.8 | 27.1 | 49.5 | 56.9 | 36.3 | 61.5 | |
| BaseMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 73.3 | 42.9 | 27.2 | 49.7 | 56.8 | 36.4 | 61.7 | |
| IP-MergingMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, T=0.32026.01 | 73.2 | 41.4 | 27.5 | 49.3 | 56.5 | 34.6 | 62.9 | |
| DAREMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 73.1 | 41.2 | 25.9 | 49.2 | 56.1 | 35.9 | 63 | |
| IP-MergingMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 72.3 | 42.1 | 30.3 | 52.6 | 59.8 | 42.1 | 68.9 | |
| IP-MergingMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, T=0.22026.01 | 72.2 | 40.5 | 24.9 | 48 | 54.9 | 33.8 | 61.9 | |
| BaseMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 72 | 42.3 | 29.5 | 52.2 | 60.7 | 39.5 | 68.9 | |
| TAMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, lambda=0.12026.01 | 71.4 | 40 | 25.7 | 49 | 56.6 | 37.8 | 62.2 | |
| FRISMMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B2026.01 | 69.9 | 42.1 | 27.2 | 49 | 56.3 | 37.5 | 61.3 | |
| DAREMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 69.6 | 40.2 | 28.9 | 50.5 | 57 | 39.8 | 67.2 | |
| Ties-MergingMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B2026.01 | 69 | 41.8 | 25.5 | 47.5 | 55.2 | 33.9 | 59.8 | |
| DAREMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B2026.01 | 68.9 | 41.9 | 27.2 | 47.9 | 54.5 | 34.8 | 60.1 | |
| TAMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 68.2 | 39.2 | 26.4 | 49.2 | 56.1 | 39.7 | 65.8 | |
| BaseMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B2026.01 | 68.1 | 41.2 | 25.6 | 47.4 | 55.1 | 34.6 | 60.1 | |
| TAMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B, lambda=0.12026.01 | 67.6 | 40.3 | 25.9 | 47.2 | 56.4 | 34.7 | 58.5 | |
| Ties-MergingMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 67.4 | 36.8 | 24.3 | 46 | 52.6 | 35.8 | 59.3 | |
| IP-MergingMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B2026.01 | 67.4 | 39.1 | 25.1 | 47.3 | 55.9 | 34.6 | 61.4 | |
| TAMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B, lambda=0.152026.01 | 67.2 | 38.2 | 23.6 | 46.1 | 53.4 | 34.9 | 59.2 | |
| Ties-MergingMerged Models=InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B2026.01 | 66 | 38.6 | 25.5 | 47.6 | 55.1 | 37.5 | 62.6 | |
| TAMerged Models=Qwen2.5-VL-7B-Instruct + OpenThinker3-7B, lambda=0.22026.01 | 63.5 | 38.4 | 21.9 | 45.3 | 54.6 | 35.2 | 58.1 | |
| TAMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, lambda=0.152026.01 | 34.7 | 19.8 | 13.2 | 28.6 | 38.7 | 23.7 | 41.7 | |
| FRISMMerging Setup=LLAVA-Next-8B with DeepThought-8B2026.01 | 32.1 | 12.8 | 10.9 | 18.6 | — | — | — | |
| TAMerged Models=ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B, lambda=0.22026.01 | 31.7 | 14.9 | 10.8 | 25.9 | 39.2 | 19 | 39.5 | |
| TAMerging Setup=LLAVA-Next-8B with DeepThought-8B, Method Type=Task Arithmetic2026.01 | 31.4 | 11.4 | 10.9 | 17.9 | — | — | — | |
| IP-MergingMerging Setup=LLAVA-Next-8B with DeepThought-8B2026.01 | 29.8 | 9.8 | 9.9 | 16.5 | — | — | — | |
| BaseBase Model=LLAVA-Next-8B2026.01 | 29 | 10.7 | 9.5 | 16.4 | — | — | — |