Clinical Question Answering on GPQA Bio
92.6AccuracySAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAGBackbone=Qwen-4B each, Optimization=w/ GRPO2026.02 | 92.6 | 7 | |
| SAGBackbone=Llama-3B each, Optimization=w/ GRPO2026.02 | 88.6 | 6.3 | |
| SAGBackbone=Qwen-4B each, Optimization=w/ CTDE2026.02 | 87.3 | 5.4 | |
| SAGBackbone=Llama-3B each, Optimization=w/ CTDE2026.02 | 79.4 | 10.2 | |
| SAGBackbone=Qwen-4B each, Optimization=Pre-trained2026.02 | 73.3 | 17.9 | |
| Single giant LLMBackbone=Qwen-72B, Optimization=w/ PPO2026.02 | 72 | 19.3 | |
| SAGBackbone=Llama-3B each, Optimization=Pre-trained2026.02 | 70.1 | 19.7 | |
| Single giant LLMBackbone=Qwen-72B, Optimization=w/ DPO2026.02 | 67.3 | 15.1 | |
| Single giant LLMBackbone=Llama-70B, Optimization=w/ DPO2026.02 | 62 | 30.7 | |
| Single giant LLMBackbone=Llama-70B, Optimization=w/ PPO2026.02 | 61.3 | 18.8 | |
| Me-LLaMAModel Type=Clinical specialist2026.02 | 51.2 | 26.9 | |
| Single giant LLMBackbone=Llama-70B, Optimization=Pre-trained2026.02 | 43.6 | 17.4 | |
| Single giant LLMBackbone=Qwen-72B, Optimization=Pre-trained2026.02 | 41.1 | 30.9 | |
| MeditronModel Type=Clinical specialist2026.02 | 20.4 | 28.3 |