List your company

Clinical Question Answering on GPQA Bio

92.6AccuracySAG
17.51237.00656.575.994Feb 8, 2026

Evaluation results

MethodLinks
SAGBackbone=Qwen-4B each, Optimization=w/ GRPO2026.0292.67
SAGBackbone=Llama-3B each, Optimization=w/ GRPO2026.0288.66.3
SAGBackbone=Qwen-4B each, Optimization=w/ CTDE2026.0287.35.4
SAGBackbone=Llama-3B each, Optimization=w/ CTDE2026.0279.410.2
SAGBackbone=Qwen-4B each, Optimization=Pre-trained2026.0273.317.9
Single giant LLMBackbone=Qwen-72B, Optimization=w/ PPO2026.027219.3
SAGBackbone=Llama-3B each, Optimization=Pre-trained2026.0270.119.7
Single giant LLMBackbone=Qwen-72B, Optimization=w/ DPO2026.0267.315.1
Single giant LLMBackbone=Llama-70B, Optimization=w/ DPO2026.026230.7
Single giant LLMBackbone=Llama-70B, Optimization=w/ PPO2026.0261.318.8
Me-LLaMAModel Type=Clinical specialist2026.0251.226.9
Single giant LLMBackbone=Llama-70B, Optimization=Pre-trained2026.0243.617.4
Single giant LLMBackbone=Qwen-72B, Optimization=Pre-trained2026.0241.130.9
MeditronModel Type=Clinical specialist2026.0220.428.3