Mathematical Reasoning on MATH-Vision
30.4AccuracyGPT4o
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT4oAccess=Close-source2026.02 | 30.4 | — | — | — | — | |
| R1-Onevision-7BParameters=7B2026.02 | 29.9 | — | — | — | — | |
| Semantic-back-7BParameters=7B2026.02 | 27.57 | — | — | — | — | |
| DAPO + AT-RLModel=Qwen2.5-VL-7B2026.02 | 27.4 | — | — | — | — | |
| OpenVLThinker-7BParameters=7B2026.02 | 25.59 | — | — | — | — | |
| DAPOModel=Qwen2.5-VL-7B2026.02 | 25.3 | — | — | — | — | |
| SAYO-Qwen-8BBase Model=Qwen3-VL-8B2026.02 | 25.26 | — | — | — | — | |
| GSPOModel=Qwen2.5-VL-7B2026.02 | 25.15 | — | — | — | — | |
| GRPOModel=Qwen2.5-VL-7B2026.02 | 24.35 | — | — | — | — | |
| Uni-MuMER-LLAVAArchitecture=LLaVA2025.05 | 24.34 | — | — | — | — | |
| Uni-MuMER2025.05 | 24.01 | — | — | — | — | |
| Qwen3-VL-30B-A3BParameters=30B-A3B2026.02 | 22.93 | — | — | — | — | |
| DAPO + AT-RLModel=Qwen2.5-VL-3B2026.02 | 22.8 | — | — | — | — | |
| Zero-shotModel=Qwen2.5-VL-7B2026.02 | 22.7 | — | — | — | — | |
| SAYO-Qwen-4BBase Model=Qwen3-VL-4B2026.02 | 22.47 | — | — | — | — | |
| Qwen3-VL-8BParameters=8B2026.02 | 22.2 | — | — | — | — | |
| ViGoRL2026.02 | 22.07 | — | — | — | — | |
| NoisyRollout-7BParameters=7B2026.02 | 22.01 | — | — | — | — | |
| Qwen3-VL-4BParameters=4B2026.02 | 21.64 | — | — | — | — | |
| FT-RLModel=Qwen2.5-VL-7B2026.02 | 21.5 | — | — | — | — | |
| Qwen2.5-VL-3B-InstructParameters=3B, Mode=Instruct2025.05 | 21.38 | — | — | — | — | |
| DAPOModel=Qwen2.5-VL-3B2026.02 | 20.6 | — | — | — | — | |
| GSPOModel=Qwen2.5-VL-3B2026.02 | 20.45 | — | — | — | — | |
| GRPOModel=Qwen2.5-VL-3B2026.02 | 19.65 | — | — | — | — | |
| SWUDImode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 18.42 | — | — | — | — | |
| InternVL2.5-Instructmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 18.09 | — | — | — | — | |
| TIES w/ DAREmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 18.09 | — | — | — | — | |
| Iso-Cmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 18.09 | — | — | — | — | |
| Zero-shotModel=Qwen2.5-VL-3B2026.02 | 17.96 | — | — | — | — | |
| Weight Averagemode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 17.76 | — | — | — | — | |
| TSV Mergingmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 17.76 | — | — | — | — | |
| Kimi-VL-16BParameters=16B2026.02 | 17.5 | — | — | — | — | |
| Task Arithmeticmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 17.43 | — | — | — | — | |
| Mixture Trainingmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 17.11 | — | — | — | — | |
| InternVL3.5-30B-A3BParameters=30B-A3B2026.02 | 17.07 | — | — | — | — | |
| TIES Mergingmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 16.78 | — | — | — | — | |
| SWUDI-Amode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 16.45 | — | — | — | — | |
| TA w/ DAREmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 15.79 | — | — | — | — | |
| OptMergemode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 15.79 | — | — | — | — | |
| WUDI Mergingmode=Full fine-tuning, backbone=InternVL2.5-1B2026.06 | 15.31 | — | — | — | — | |
| FT-RLModel=Qwen2.5-VL-3B2026.02 | 15.2 | — | — | — | — | |
| InternVL3.5-14BParameters=14B2026.02 | 13.85 | — | — | — | — | |
| InternVL3.5-8BParameters=8B2026.02 | 13.82 | — | — | — | — | |
| SAYO-InternVL-8BBase Model=InternVL3.5-8B2026.02 | 13.62 | — | — | — | — | |
| InternVL3.5-38BParameters=38B2026.02 | 13.36 | — | — | — | — | |
| AHD (Ours)2026.04 | — | — | — | 10 | 2.16 | |
| Aquila-VL-2Bparameters=2B2024.12 | — | 21.1 | 18.4 | — | — | |
| Claude-3.5 Sonnet-0620Model Version=06202025.02 | — | — | — | 30.4 | — | |
| Gemini-1.5-Pro2024.12 | — | — | 19.2 | — | — | |
| GPT-4o 0513Model Version=05132025.02 | — | — | — | 32.2 | — | |
| GPT-4o-202405132024.12 | — | — | 30.4 | — | — | |
| GPT-4V2024.12 | — | — | 24 | — | — | |
| InternVL-Chat-V1.52024.12 | — | 15.8 | 15 | — | — | |
| InternVL2-1Bparameters=1B2024.12 | — | 12.2 | 11.1 | — | — | |
| InternVL2-26Bparameters=26B2024.12 | — | 23.4 | 17 | — | — | |
| InternVL2-2Bparameters=2B2024.12 | — | 15.8 | 12.1 | — | — | |
| InternVL2-40Bparameters=40B2024.12 | — | 21.4 | 16.9 | — | — | |
| InternVL2-4Bparameters=4B2024.12 | — | 17.8 | 16.5 | — | — | |
| InternVL2-8Bparameters=8B2024.12 | — | 20.4 | 18.4 | — | — | |
| InternVL2-Llama3-76Bparameters=76B, backbone=Llama32024.12 | — | 23.7 | 23.6 | — | — | |
| InternVL2.5Number of Parameters=78B2025.02 | — | — | — | 25.9 | — | |
| InternVL2.5 (Previous SoTA)reference=Chen et al. (2024d)2025.02 | — | — | — | 32.2 | — | |
| InternVL2.5-1Bparameters=1B2024.12 | — | 16.8 | 14.4 | — | — | |
| InternVL2.5-26Bparameters=26B2024.12 | — | 28 | 23.1 | — | — | |
| InternVL2.5-2Bparameters=2B2024.12 | — | 13.5 | 14.7 | — | — | |
| InternVL2.5-38Bparameters=38B2024.12 | — | 32.2 | 31.8 | — | — | |
| InternVL2.5-4Bparameters=4B2024.12 | — | 21.7 | 20.9 | — | — | |
| InternVL2.5-78Bparameters=78B2024.12 | — | 34.9 | 32.2 | — | — | |
| InternVL2.5-8Bparameters=8B2024.12 | — | 22 | 19.7 | — | — | |
| MiniCPM-V2.62024.12 | — | 16.1 | 17.5 | — | — | |
| MMaDA2026.04 | — | — | — | 9.44 | 1 | |
| Ovis1.6-Gemma2-9Bparameters=9B, backbone=Gemma22024.12 | — | — | 18.8 | — | — | |
| Phi-3.5-Vision-4Bparameters=4B2024.12 | — | 17.4 | 15.5 | — | — | |
| Qwen2-VL-2Bparameters=2B2024.12 | — | 19.7 | 12.4 | — | — | |
| Qwen2-VL-72Bparameters=72B2024.12 | — | — | 25.9 | — | — | |
| Qwen2-VL-7Bparameters=7B2024.12 | — | 22 | 16.3 | — | — | |
| Qwen2.5-VLNumber of Parameters=72B2025.02 | — | — | — | 38.1 | — | |
| Qwen2.5-VLNumber of Parameters=7B2025.02 | — | — | — | 25.1 | — | |
| Qwen2.5-VLNumber of Parameters=3B2025.02 | — | — | — | 21.2 | — |