Mathematical Reasoning on MathVista (testmini)
85.9AccuracyUniDFlow
Evaluation results
| Method | Links | ||
|---|---|---|---|
| UniDFlowParams=4B2026.02 | 85.9 | — | |
| Gemini 2.5 ProAccess=API call only2026.03 | 84.6 | — | |
| GPT-5Access=API call only2026.03 | 82.7 | — | |
| MammothModa-2Params=4B2026.02 | 81.8 | — | |
| Q3VL 8B-ThkRL=N/A, Initial Model=N/A2026.04 | 81.4 | — | |
| Gemini 2.5 FlashAccess=API call only2026.03 | 81.2 | — | |
| Keye-VL-1.5-8BAccess=Open weights only2026.03 | 81.2 | — | |
| GLM-4.1V-9BAccess=Open weights only2026.03 | 80.7 | — | |
| MiMoVL 7B-RLRL=MiMoVL, Initial Model=7B-SFT2026.04 | 80.4 | — | |
| MiniCPM-V-4.5-8BAccess=Open weights only2026.03 | 79.9 | — | |
| GPT-5 miniAccess=API call only2026.03 | 79.2 | — | |
| Vero Q3T-8BRL=Qwen3VL, Initial Model=8B Think2026.04 | 79.2 | — | |
| MudditParams=4B2026.02 | 79.1 | — | |
| Vero Mi-7BRL=MiMoVL, Initial Model=7B-SFT2026.04 | 79.1 | — | |
| Vero Q3I-8BRL=Qwen3VL, Initial Model=8B Inst2026.04 | 78.7 | — | |
| InternVL3.5-8BAccess=Open weights only2026.03 | 78.4 | — | |
| PRISM + GRPOBase Model=Qwen3-VL-8B2026.04 | 78.3 | — | |
| PRISM + DAPOBase Model=Qwen3-VL-8B2026.04 | 78.2 | — | |
| PRISM + GRPOBase Model=Qwen3-VL-4B2026.04 | 77.9 | — | |
| PRISM + GSPOBase Model=Qwen3-VL-8B2026.04 | 77.9 | — | |
| PRISM + DAPOBase Model=Qwen3-VL-4B2026.04 | 77.8 | — | |
| PRISM + GSPOBase Model=Qwen3-VL-4B2026.04 | 77.5 | — | |
| Qwen3-VL-8BAccess=Open weights only2026.03 | 77.2 | — | |
| Q3VL 8B-InsRL=N/A, Initial Model=N/A2026.04 | 77.2 | — | |
| InternVL3.5-4BAccess=Open weights only2026.03 | 77.1 | — | |
| + DAPOBase Model=Qwen3-VL-8B2026.04 | 77 | — | |
| InstructBase Model=Qwen3-VL-8B2026.04 | 76 | — | |
| + GRPOBase Model=Qwen3-VL-8B2026.04 | 75.9 | — | |
| + GSPOBase Model=Qwen3-VL-8B2026.04 | 75.9 | — | |
| EMMAParams=4B2026.02 | 75.8 | — | |
| + GRPOBase Model=Qwen3-VL-4B2026.04 | 75.7 | — | |
| + GSPOBase Model=Qwen3-VL-4B2026.04 | 75.2 | — | |
| InstructBase Model=Qwen3-VL-4B2026.04 | 74.9 | — | |
| Vero Q25-7BRL=Qwen25VL, Initial Model=7B Inst2026.04 | 74.4 | — | |
| + DAPOBase Model=Qwen3-VL-4B2026.04 | 74.3 | — | |
| Qwen3-VL-4BAccess=Open weights only2026.03 | 73.7 | — | |
| VL-RethinkerRL=Q25VL, Initial Model=7B-Ins2026.04 | 73.7 | — | |
| BAGELParams=7B2026.02 | 73.1 | — | |
| BAGELModel Architecture Style=AR Based2026.03 | 73.1 | — | |
| Claude Sonnet 4.5Access=API call only2026.03 | 73.1 | — | |
| LLaVA OV1.5-RLRL=LLaVA, Initial Model=OV1.5-Ins2026.04 | 72.3 | — | |
| + SFTBase Model=Qwen3-VL-4B2026.04 | 71.5 | — | |
| PRISMBase Model=Qwen3-VL-8B2026.04 | 71.4 | — | |
| PRISMBase Model=Qwen3-VL-4B2026.04 | 71 | — | |
| GPT-5 NanoRL=N/A, Initial Model=N/A2026.04 | 70.2 | — | |
| + SFTBase Model=Qwen3-VL-8B2026.04 | 70.2 | — | |
| PLM-HoneyBee-8BModel scale=7B-8B2025.10 | 68.2 | — | |
| Q25VL 7B-InsRL=N/A, Initial Model=N/A2026.04 | 68.2 | — | |
| Eagle2.5-8BAccess=Open weights only2026.03 | 67.8 | — | |
| Qwen2.5-VL-7B-InstructModel scale=7B-8B2025.10 | 67.5 | — | |
| LLaDA-oModel Architecture Style=Diff. Based2026.03 | 66.1 | — | |
| Qwen2.5-VLParams=3B2026.02 | 62.3 | — | |
| InternVL-2.5-4BModel scale=3B-4B2025.10 | 61.8 | — | |
| InternVL-3-8B-InstructModel scale=7B-8B2025.10 | 61.8 | — | |
| InternVL-2.5-8BModel scale=7B-8B2025.10 | 61.5 | — | |
| GPT-4oZero-shot=true, Decoding Strategy=greedy2024.09 | 61.3 | — | |
| PLM-HoneyBee-3BModel scale=3B-4B2025.10 | 61.2 | — | |
| PLM-8BAccess=Fully open2026.03 | 59.9 | — | |
| LLaDA-VModel Architecture Style=Diff. Based2026.03 | 59.7 | — | |
| MolmoPoint-8BAccess=MolmoPoint2026.03 | 59.4 | — | |
| PLM-8BModel scale=7B-8B2025.10 | 59.3 | — | |
| PLM-3BAccess=Fully open2026.03 | 59.1 | — | |
| Qwen2.5-VL-3B-InstructModel scale=3B-4B2025.10 | 58.9 | — | |
| Molmo2-8BAccess=Fully open2026.03 | 58.9 | — | |
| Gemini-1.5-ProZero-shot=true, Decoding Strategy=greedy2024.09 | 57.7 | — | |
| PLM-3BModel scale=3B-4B2025.10 | 57.2 | — | |
| LaViDa-OModel Architecture Style=Diff. Based2026.03 | 56.9 | — | |
| Molmo2-4BAccess=Fully open2026.03 | 56.7 | — | |
| MM1.5-30BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 55.6 | — | |
| MolmoPoint-8B-O-7BAccess=Fully open2026.03 | 54.2 | — | |
| PLM-HoneyBee-1BModel scale=1B2025.10 | 53.7 | — | |
| Mo2-O 7BRL=SFT Only, Initial Model=N/A2026.04 | 53.6 | — | |
| InternVL 1.5#param=26B, open-source=false2024.04 | 53.5 | — | |
| Cambrian-34BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 53.2 | — | |
| Gemini Ultra 1.0open-source=false2024.04 | 53 | — | |
| Grok-1.5Vopen-source=false2024.04 | 52.8 | — | |
| Gemini Pro 1.5open-source=false2024.04 | 52.1 | — | |
| Idefics2-8BModel Scale=7B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 51.4 | — | |
| Qwen-VL-Maxopen-source=false2024.04 | 51 | — | |
| Claude-3 Opusopen-source=false2024.04 | 50.5 | — | |
| GPT-4Vopen-source=false2024.04 | 49.9 | — | |
| GPT-4VZero-shot=true, Decoding Strategy=greedy2024.09 | 48.7 | — | |
| PLM-1BModel scale=1B2025.10 | 48.6 | — | |
| Claude-3 Sonnetopen-source=false2024.04 | 47.9 | — | |
| InternVL 1.2#param=40B, open-source=false2024.04 | 47.7 | — | |
| MM1.5-7BModel Scale=7B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 47.6 | — | |
| MM1.5-3B-MoEModel Scale=3B, Architecture=MoE, Zero-shot=true, Decoding Strategy=greedy2024.09 | 46.9 | — | |
| LLaVA-NeXT#param=35B, open-source=false2024.04 | 46.5 | — | |
| LLaVA-NeXT-34BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 46.5 | — | |
| Claude-3 Haikuopen-source=false2024.04 | 46.4 | — | |
| InternVL2-2BModel Scale=1B, Zero-shot=true, Decoding Strategy=beam search2024.09 | 46 | — | |
| Gemini Pro 1.0open-source=false2024.04 | 45.2 | — | |
| Step-1V#param=100B, open-source=false2024.04 | 44.8 | — | |
| LaViDa-LModel Architecture Style=Diff. Based2026.03 | 44.8 | — | |
| Phi-3-Vision-4BModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 44.5 | — | |
| MM1.5-3BModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 44.4 | — | |
| InternVL-2.5-1BModel scale=1B2025.10 | 44.2 | — | |
| Qwen-VL-Plusopen-source=false2024.04 | 43.3 | — | |
| Mini-Gemini#param=35B, open-source=false2024.04 | 43.3 | — | |
| MM1.5-1B-MoEModel Scale=1B, Architecture=MoE, Zero-shot=true, Decoding Strategy=greedy2024.09 | 42.9 | — |