Radiology Report Generation on MIMIC-CXR (RaTE, Semb, RadCliQ^-1)
52.6RaTEEviAgent
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EviAgentbackbone=Qwen3-VL-8B2026.03 | 52.6 | 43.6 | 76.6 | |
| MedGemma-4B-ITParameters=4B2026.03 | 52.4 | 29.2 | 62.9 | |
| Lingshu-7BParameters=7B2026.03 | 52.1 | 30 | 69.2 | |
| Gemini-2.5-Flash2026.03 | 50.3 | 29.7 | 59.4 | |
| GPT-5.12026.03 | 49.5 | 28 | 65.6 | |
| Claude 4.5 Sonnet2026.03 | 49.1 | 23.9 | 64.7 | |
| HuatuoGPT-V-7BParameters=7B2026.03 | 48.9 | 20 | 48.2 | |
| Qwen3-VL-8BParameters=8B2026.03 | 48.9 | 26.1 | 64.2 | |
| InternVL3-8BParameters=8B2026.03 | 48.2 | 21.5 | 55.1 | |
| InternVL2.5-8BParameters=8B2026.03 | 47 | 21 | 56.2 | |
| Qwen2.5-VL-7BParameters=7B2026.03 | 47 | 18.4 | 55.1 | |
| BiMediX2-8BParameters=8B2026.03 | 44.4 | 17.7 | 53 | |
| LLaVA-Med-7BParameters=7B2026.03 | 12.8 | 18.3 | 52.9 |