Medical Visual Question Answering on MicroVQA
67.1Overall AccuracyQwen3-VL-8B-Instruct
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen3-VL-8B-InstructScale=8B, Mode=Two-Pass2026.05 | 67.1 | — | — | — | 67.9 | 68.4 | 63.5 | |
| GPT-o4-miniModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 65 | 68.75 | 67.78 | 54.35 | — | — | — | |
| Qwen3-VL-4B-InstructScale=4B, Mode=Two-Pass2026.05 | 64.3 | — | — | — | 62.9 | 66.6 | 63 | |
| InternVL3.5-8B-InstructScale=8B, Mode=Two-Pass2026.05 | 63.4 | — | — | — | 64 | 63.5 | 62.2 | |
| Gemini-2.5-proModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 60.9 | 61.2 | 62.6 | 57.4 | — | — | — | |
| GPT-5Model size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 60.8 | 62 | 61.2 | 58.3 | — | — | — | |
| GPT-5Mode=Baseline2026.05 | 59.4 | — | — | — | 58.9 | 53.9 | 63.3 | |
| Qwen3-VL-2B-InstructScale=2B, Mode=Two-Pass2026.05 | 59.4 | — | — | — | 60 | 57.4 | 61.7 | |
| o3Mode=Baseline2026.05 | 59.3 | — | — | — | 60.5 | 53.5 | 61.5 | |
| Qwen3-VL-8B-InstructScale=8B, Mode=Single-Pass2026.05 | 58.9 | — | — | — | 58.3 | 60.5 | 57.4 | |
| GPT-5-miniModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 57.1 | 58.7 | 57.61 | 53.51 | — | — | — | |
| Claude-4.5-sonnetModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 56.33 | 55.6 | 57.91 | 54.83 | — | — | — | |
| InternVL3.5-2B-InstructScale=2B, Mode=Two-Pass2026.05 | 56.3 | — | — | — | 53.3 | 55.4 | 63.5 | |
| o4-miniMode=Baseline2026.05 | 55.6 | — | — | — | 56.1 | 50.4 | 57.9 | |
| InternVL3.5-8B-InstructScale=8B, Mode=Single-Pass2026.05 | 54.8 | — | — | — | 55 | 55.6 | 53 | |
| Qwen3-VL-4B-InstructScale=4B, Mode=Single-Pass2026.05 | 54.7 | — | — | — | 55.2 | 55.1 | 53 | |
| Claude Sonnet 4.5Mode=Baseline2026.05 | 54.4 | — | — | — | 56.4 | 49.6 | 55.1 | |
| Claude-4.5-haikuModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 53.14 | 52.81 | 51.21 | 57 | — | — | — | |
| o1Mode=Baseline2026.05 | 52.8 | — | — | — | 50.2 | 53 | 55.4 | |
| Qwen3-VL-2B-InstructScale=2B, Mode=Single-Pass2026.05 | 52.8 | — | — | — | 51.9 | 51.8 | 56.1 | |
| InternVL3.5-2B-InstructScale=2B, Mode=Single-Pass2026.05 | 51.3 | — | — | — | 49.5 | 49.2 | 58.3 | |
| InternVL3.5-4B-InstructScale=4B, Mode=Two-Pass2026.05 | 49.8 | — | — | — | 48.1 | 51 | 50.9 | |
| Qwen3-VL-8B-InstructScale=8B, Mode=Baseline2026.05 | 48.8 | — | — | — | 48.6 | 49.2 | 48.3 | |
| Gemini-2.5-flashModel size=Proprietary, Prompting protocol=vanilla prompt2026.01 | 47 | 50 | 44.5 | 46.5 | — | — | — | |
| Qwen3-VL-4B-InstructScale=4B, Mode=Baseline2026.05 | 46.4 | — | — | — | 46.4 | 47.4 | 44.8 | |
| Qwen3-VL-2B-InstructScale=2B, Mode=Baseline2026.05 | 45.9 | — | — | — | 42.4 | 44.9 | 53.9 | |
| InternVL3.5-4B-InstructScale=4B, Mode=Baseline2026.05 | 45.6 | — | — | — | 43.1 | 45.4 | 50.4 | |
| InternVL3.5-2B-InstructScale=2B, Mode=Baseline2026.05 | 44.5 | — | — | — | 42.9 | 43.4 | 49.6 | |
| InternVL3.5-4B-InstructScale=4B, Mode=Single-Pass2026.05 | 44.4 | — | — | — | 43.8 | 45.9 | 43 | |
| GRPO w/o ToolsModel size=7-13B parameters, Tool-access setting=removed, Reasoning-stage setting=GRPO2026.01 | 43 | 40.63 | 44.44 | 43.48 | — | — | — | |
| MEDVISTA-R1Model size=7-13B parameters2026.01 | 43 | 42.2 | 37.8 | 54.41 | — | — | — | |
| Cold-start w/o ToolsModel size=7-13B parameters, Tool-access setting=removed, SFT-stage setting=Cold-start2026.01 | 42.5 | 43.75 | 43.33 | 39.1 | — | — | — | |
| InternVL3.5-8B-InstructScale=8B, Mode=Baseline2026.05 | 41.8 | — | — | — | 38.1 | 42.6 | 47.4 | |
| Direct GRPO w/o cold-startModel size=7-13B parameters, SFT-stage setting=removed, Reasoning-stage setting=Direct GRPO2026.01 | 41 | 43.75 | 38.89 | 41.3 | — | — | — | |
| MEDVISTA-R1Model size=< 7B parameters2026.01 | 39.5 | 37.5 | 37.78 | 45.65 | — | — | — | |
| Cold-start w/o ReasoningModel size=7-13B parameters, Reasoning-stage setting=removed, SFT-stage setting=Cold-start2026.01 | 39 | 39.1 | 37.8 | 41.3 | — | — | — | |
| Direct GRPO w/o cold-startModel size=< 7B parameters, SFT-stage setting=removed, Reasoning-stage setting=Direct GRPO2026.01 | 38.5 | 35.9 | 36.67 | 43.48 | — | — | — | |
| GRPO w/o ToolsModel size=< 7B parameters, Tool-access setting=removed, Reasoning-stage setting=GRPO2026.01 | 37 | 34.38 | 37.78 | 39.13 | — | — | — | |
| Cold-start w/o ReasoningModel size=< 7B parameters, Reasoning-stage setting=removed, SFT-stage setting=Cold-start2026.01 | 37 | 35.9 | 38.9 | 34.8 | — | — | — | |
| MEDVISTAGYM (Qwen3vl-8B)Model size=7-13B parameters, Tool-access setting=enabled2026.01 | 36.28 | 40.62 | 35.61 | 32.63 | — | — | — | |
| Qwen3vl-8BModel size=7-13B parameters2026.01 | 35.9 | 38.8 | 31.9 | 38.3 | — | — | — | |
| Cold-start w/o ToolsModel size=< 7B parameters, Tool-access setting=removed, SFT-stage setting=Cold-start2026.01 | 35.5 | 39.1 | 41.11 | 34.78 | — | — | — | |
| MEDVISTAGYM (InternVL3-8B)Model size=7-13B parameters, Tool-access setting=enabled2026.01 | 35 | 39.06 | 32.2 | 34.8 | — | — | — | |
| Qwen2.5vl-7BModel size=7-13B parameters2026.01 | 33 | 34.38 | 31.11 | 34.78 | — | — | — | |
| InternVL3-8BModel size=7-13B parameters2026.01 | 31.5 | 31.9 | 26.7 | 39.6 | — | — | — | |
| MEDVISTAGYM (InternVL3-2B)Model size=< 7B parameters, Tool-access setting=enabled2026.01 | 31 | 34.4 | 25.6 | 37 | — | — | — | |
| Internvl3-2BModel size=< 7B parameters2026.01 | 30.48 | 30 | 28.57 | 34.86 | — | — | — | |
| LLaVA-Med-7BModel size=7-13B parameters2026.01 | 16.51 | 12.2 | 17.6 | 21.7 | — | — | — |