Rare-disease diagnosis on CROSSRARE-BENCH RareBench cohort 1.0 (test)
75AccuracyMedLatentDx
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MedLatentDxModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 75 | 64 | 77 | |
| MedLatentDxModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 72 | 61 | 75 | |
| Raw KVModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 63 | 45 | 75 | |
| Struct. RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 61 | 45 | 70 | |
| LatentMASModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 61 | 46 | 72 | |
| Global RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 58 | 41 | 68 | |
| Dist. RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 56 | 44 | 72 | |
| Single RAGModel=Qwen3-4B, Collaboration Category=Non-collaborative2026.06 | 53 | 37 | 58 | |
| TextMASModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 53 | 37 | 61 | |
| Struct. RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 53 | 35 | 63 | |
| Global RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 46 | 41 | 53 | |
| Dist. RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 44 | 40 | 60 | |
| Single RAGModel=Llama-3.2-3B, Collaboration Category=Non-collaborative2026.06 | 37 | 36 | 53 | |
| TextMASModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 33 | 24 | 49 | |
| Raw KVModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 28 | 24 | 46 | |
| LatentMASModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 23 | 20 | 44 | |
| Zero-shotModel=Qwen3-4B, Collaboration Category=Non-collaborative2026.06 | 11 | 8 | 30 | |
| Zero-shotModel=Llama-3.2-3B, Collaboration Category=Non-collaborative2026.06 | 2 | 1 | 21 | |
| MedLatentDxModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 0.73 | 0.72 | 0.78 | |
| MedLatentDxHost Model=MediPhi, Communication Protocol=Latent communication2026.06 | 0.67 | 0.49 | 0.73 | |
| Global RAGHost Model=MediPhi, Communication Protocol=Text / structured sharing2026.06 | 0.65 | 0.49 | 0.77 | |
| Struct. RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 0.64 | 0.56 | 0.73 | |
| MedLatentDxModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 0.64 | 0.63 | 0.73 | |
| Global RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 0.62 | 0.55 | 0.72 | |
| Dist. RAGModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 0.61 | 0.54 | 0.72 | |
| LatentMASHost Model=MediPhi, Communication Protocol=Latent communication2026.06 | 0.61 | 0.42 | 0.7 | |
| Dist. RAGHost Model=MediPhi, Communication Protocol=Text / structured sharing2026.06 | 0.6 | 0.45 | 0.67 | |
| Struct. RAGHost Model=MediPhi, Communication Protocol=Text / structured sharing2026.06 | 0.6 | 0.43 | 0.7 | |
| Raw KVHost Model=MediPhi, Communication Protocol=Latent communication2026.06 | 0.6 | 0.41 | 0.7 | |
| Single RAGModel=Qwen3-4B, Collaboration Category=Non-collaborative2026.06 | 0.58 | 0.51 | 0.7 | |
| LatentMASModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 0.57 | 0.51 | 0.71 | |
| TextMASModel=Qwen3-4B, Collaboration Category=Text / structured sharing2026.06 | 0.56 | 0.48 | 0.72 | |
| Raw KVModel=Qwen3-4B, Collaboration Category=Latent communication2026.06 | 0.56 | 0.49 | 0.7 | |
| Global RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 0.5 | 0.44 | 0.66 | |
| Dist. RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 0.47 | 0.4 | 0.64 | |
| Single RAGHost Model=MediPhi, Communication Protocol=Non-collaborative2026.06 | 0.47 | 0.32 | 0.56 | |
| TextMASHost Model=MediPhi, Communication Protocol=Text / structured sharing2026.06 | 0.46 | 0.27 | 0.65 | |
| Single RAGModel=Llama-3.2-3B, Collaboration Category=Non-collaborative2026.06 | 0.43 | 0.37 | 0.62 | |
| Struct. RAGModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 0.43 | 0.37 | 0.69 | |
| Raw KVModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 0.4 | 0.35 | 0.64 | |
| LatentMASModel=Llama-3.2-3B, Collaboration Category=Latent communication2026.06 | 0.38 | 0.35 | 0.62 | |
| TextMASModel=Llama-3.2-3B, Collaboration Category=Text / structured sharing2026.06 | 0.31 | 0.25 | 0.47 | |
| Zero-shotHost Model=MediPhi, Communication Protocol=Non-collaborative2026.06 | 0.14 | 0.12 | 0.25 | |
| Zero-shotModel=Qwen3-4B, Collaboration Category=Non-collaborative2026.06 | 0.01 | 0.01 | 0.18 | |
| Zero-shotModel=Llama-3.2-3B, Collaboration Category=Non-collaborative2026.06 | 0 | 0 | 0.1 |