Fundus Reading on Omni-Fundus
97.7AccuracyLingshu-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Lingshu-7BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-7B, Post-training Data Scale=5.1M, Protocol=Medical MLLMs (SFT)2026.04 | 97.7 | |
| Fundus-R1-7BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-7B, Post-training Data Scale=80.1K, Protocol=Fundus-reading MLLMs (RLVR)2026.04 | 91.1 | |
| Fundus-R1-3BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-3B, Post-training Data Scale=80.1K, Protocol=Fundus-reading MLLMs (RLVR)2026.04 | 79.8 | |
| Med-R1-fundusVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-3B, Post-training Data Scale=undisclosed, Protocol=Fundus-reading MLLMs (RLVR)2026.04 | 79.2 | |
| FundusExpert-8BVision Encoder=InternViT-300M, LLM=InternLM2.5-7B, Post-training Data Scale=200K, Protocol=Fundus-reading MLLMs (SFT)2026.04 | 78.4 | |
| HuatuoGPT-Vision-7BVision Encoder=OpenAI CLIP ViT-L/14, LLM=Qwen2-7B, Post-training Data Scale=1.3M, Protocol=Medical MLLMs (SFT)2026.04 | 69.6 | |
| Qwen2.5-VL-7BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-7B, Post-training Data Scale=–, Protocol=Generic MLLMs2026.04 | 68.1 | |
| Qwen2.5-VL-3BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-3B, Post-training Data Scale=–, Protocol=Generic MLLMs2026.04 | 66.4 | |
| HealthGPT-M3-7BVision Encoder=CLIP ViT-L/14-336, LLM=Phi-3-mini, Post-training Data Scale=1.6M, Protocol=Medical MLLMs (SFT)2026.04 | 64 | |
| OphthaReason-QwenVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-3B, Post-training Data Scale=121.5K, Protocol=Fundus-reading MLLMs (RLVR)2026.04 | 62.4 | |
| QoQ-Med-7BVision Encoder=Qwen2.5-ViT, LLM=Qwen2.5-7B, Post-training Data Scale=2.6M, Protocol=Medical MLLMs (RLVR)2026.04 | 57.9 | |
| OphthaReason-InternVision Encoder=InternViT-300M, LLM=Qwen2.5-1.5B, Post-training Data Scale=121.5K, Protocol=Fundus-reading MLLMs (RLVR)2026.04 | 57.7 | |
| MedVLM-R1Vision Encoder=Qwen2-ViT, LLM=Qwen2-2B, Post-training Data Scale=600, Protocol=Medical MLLMs (RLVR)2026.04 | 56.3 | |
| InternVL2.5-8BVision Encoder=InternViT-300M, LLM=InternLM2.5-7B, Post-training Data Scale=–, Protocol=Generic MLLMs2026.04 | 55.6 |