Multimodal Reasoning on ScienceQA
97.85Average AccuracyMG2-RAG
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| MG2-RAGLLM/MLLM=Qwen3.5-27B2026.04 | 97.85 | 98.45 | 95.28 | 98.73 | 98.44 | 96.43 | 98.68 | 98.42 | 96.84 | — | |
| Qwen3.5-27BLLM/MLLM=Qwen3.5-27B2026.04 | 97.34 | 98.27 | 95.16 | 97.18 | 97.9 | 96.43 | 97.49 | 97.76 | 96.57 | — | |
| MG2-RAGLLM/MLLM=GPT-5.22026.04 | 97 | 97.78 | 94.65 | 97.06 | 97.93 | 96.24 | 97.05 | 97.33 | 96.43 | — | |
| GPT-5.2LLM/MLLM=GPT-5.22026.04 | 96.3 | 97.6 | 93.05 | 95.96 | 97.73 | 95.35 | 96.17 | 96.54 | 95.88 | — | |
| Gemini-3.1-ProLLM/MLLM=Gemini-3.1-Pro2026.04 | 92.9 | 93.53 | 94.12 | 90.81 | 92.98 | 94.25 | 90.27 | 94.03 | 90.93 | — | |
| MG2-RAGLLM/MLLM=Qwen2.5-VL-7B2026.04 | 89.25 | 90.81 | 87.96 | 87.09 | 89.93 | 86.61 | 88.92 | 90.64 | 86.75 | — | |
| HumanLLM/MLLM=-2026.04 | 88.4 | 90.23 | 84.97 | 87.48 | 89.6 | 87.5 | 88.1 | 91.59 | 82.42 | — | |
| MG2-RAGLLM/MLLM=Qwen2.5-72B2026.04 | 87.62 | 88.37 | 81.21 | 91.27 | 87 | 78.28 | 94.01 | 88.84 | 85.43 | — | |
| VISTA-SFTModel Variant=Qwen2.5-VL-7B2026.05 | 87.58 | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BLLM/MLLM=Qwen2.5-VL-7B2026.04 | 87.5 | 88.77 | 88.86 | 83.82 | 87.39 | 88.05 | 86.06 | 89.94 | 83.12 | — | |
| LaReBase VLM=Qwen3-VL-4B-Instruct2025.11 | 87.5 | — | — | — | — | — | — | — | — | 65.2 | |
| R3VModel Variant=Qwen2.5-VL-7B2026.05 | 86.83 | — | — | — | — | — | — | — | — | — | |
| STaR+Model Variant=Qwen2.5-VL-7B2026.05 | 85.73 | — | — | — | — | — | — | — | — | — | |
| SFT-OracleModel Variant=Qwen2.5-VL-7B2026.05 | 85.49 | — | — | — | — | — | — | — | — | — | |
| VaLiKLLM/MLLM=Qwen2.5-72B2026.04 | 84.77 | 85.61 | 75.93 | 90.27 | 84.4 | 74.17 | 92.33 | 85.79 | 82.98 | — | |
| MG2-RAGLLM/MLLM=Qwen2.5-7B2026.04 | 84.53 | 85.17 | 81.33 | 85.82 | 83.43 | 76.1 | 89.41 | 86.12 | 81.67 | — | |
| STaRModel Variant=Qwen2.5-VL-7B2026.05 | 84.39 | — | — | — | — | — | — | — | — | — | |
| CoTLLM/MLLM=GPT-42026.04 | 83.99 | 85.48 | 72.44 | 90.27 | 82.65 | 71.49 | 92.89 | 86.66 | 79.04 | — | |
| CCoTBase VLM=Qwen3-VL-4B-Instruct2025.11 | 83.4 | — | — | — | — | — | — | — | — | 240.2 | |
| VaLiKLLM/MLLM=Qwen2.5-7B2026.04 | 83.16 | 84.15 | 75.14 | 87.64 | 82.99 | 73.18 | 89.69 | 84.4 | 80.95 | — | |
| MM-CoTBase VLM=Qwen3-VL-4B-Instruct2025.11 | 82.7 | — | — | — | — | — | — | — | — | 203.1 | |
| LaReBase VLM=Qwen3-VL-2B-Instruct2025.11 | 82.5 | — | — | — | — | — | — | — | — | 55 | |
| VISTA-SFTModel Variant=Qwen2.5-VL-3B2026.05 | 82.29 | — | — | — | — | — | — | — | — | — | |
| SFT-OracleModel Variant=Qwen2.5-VL-3B2026.05 | 82 | — | — | — | — | — | — | — | — | — | |
| ICoTBase VLM=Qwen3-VL-4B-Instruct2025.11 | 81.7 | — | — | — | — | — | — | — | — | 172.6 | |
| SFT-SeedModel Variant=Qwen2.5-VL-7B2026.05 | 81.55 | — | — | — | — | — | — | — | — | — | |
| R3VModel Variant=Qwen2.5-VL-3B2026.05 | 80.81 | — | — | — | — | — | — | — | — | — | |
| CoF*Base VLM=Qwen3-VL-4B-Instruct2025.11 | 80.6 | — | — | — | — | — | — | — | — | 205.5 | |
| LVR*Base VLM=Qwen3-VL-4B-Instruct2025.11 | 80.3 | — | — | — | — | — | — | — | — | 101.4 | |
| MM-CoTBase VLM=Qwen3-VL-2B-Instruct2025.11 | 80.1 | — | — | — | — | — | — | — | — | 195.7 | |
| ChameleonLLM/MLLM=ChatGPT2026.04 | 79.93 | 81.62 | 70.64 | 84 | 79.77 | 70.8 | 86.62 | 81.86 | 76.53 | — | |
| CCoTBase VLM=Qwen3-VL-2B-Instruct2025.11 | 79.9 | — | — | — | — | — | — | — | — | 200.9 | |
| Standard CoTBase VLM=Qwen3-VL-4B-Instruct2025.11 | 79.5 | — | — | — | — | — | — | — | — | 186.2 | |
| LVR*Base VLM=Qwen3-VL-2B-Instruct2025.11 | 79.2 | — | — | — | — | — | — | — | — | 93.8 | |
| MCOUTBase VLM=Qwen3-VL-4B-Instruct2025.11 | 78.8 | — | — | — | — | — | — | — | — | 97.5 | |
| STaR+Model Variant=Qwen2.5-VL-3B2026.05 | 78.6 | — | — | — | — | — | — | — | — | — | |
| MMGraphRAGLLM/MLLM=Qwen2.5-VL-7B2026.04 | 78.21 | 81.08 | 68.62 | 80.09 | 79.52 | 68.96 | 83.69 | 80.87 | 73.43 | — | |
| STaRModel Variant=Qwen2.5-VL-3B2026.05 | 77.49 | — | — | — | — | — | — | — | — | — | |
| ICoTBase VLM=Qwen3-VL-2B-Instruct2025.11 | 77 | — | — | — | — | — | — | — | — | 174.2 | |
| MCOUTBase VLM=Qwen3-VL-2B-Instruct2025.11 | 76.8 | — | — | — | — | — | — | — | — | 104.3 | |
| Visual GenomeLLM/MLLM=Qwen2.5-7B2026.04 | 75.08 | 76.78 | 67.04 | 78.09 | 74.05 | 66.19 | 79.72 | 78.08 | — | — | |
| SFT-SeedModel Variant=Qwen2.5-VL-3B2026.05 | 74.66 | — | — | — | — | — | — | — | — | — | |
| Standard CoTBase VLM=Qwen3-VL-2B-Instruct2025.11 | 74.4 | — | — | — | — | — | — | — | — | 182.5 | |
| MMKGLLM/MLLM=Qwen2.5-7B2026.04 | 73.47 | 73.98 | 66.37 | 78.18 | 71.65 | 64.3 | 79.65 | 76.51 | 68.03 | — | |
| CoF*Base VLM=Qwen3-VL-2B-Instruct2025.11 | 72.5 | — | — | — | — | — | — | — | — | 195.7 |