Clinical Reasoning on MEDREASON
73.4Pass@1Outcome Reward
Evaluation Results
| Method | Links | |
|---|---|---|
| Outcome RewardBackbone=Llama3.1-8B2025.10 | 73.4 | |
| Ours (Sparse)Backbone=Llama3.1-8B2025.10 | 73.1 | |
| Ours (Interval)Backbone=Llama3.1-8B2025.10 | 71.4 | |
| Ours (Dense)Backbone=Llama3.1-8B2025.10 | 69.8 | |
| Ours (Sparse)Backbone=Qwen3-4B2025.10 | 66.1 | |
| Outcome RewardBackbone=Qwen2.5-7B2025.10 | 65 | |
| Ours (Interval)Backbone=Qwen2.5-7B2025.10 | 63.8 | |
| SFTBackbone=Llama3.1-8B2025.10 | 63.5 | |
| Ours (Dense)Backbone=Qwen2.5-7B2025.10 | 58.3 | |
| SFTBackbone=Qwen3-4B2025.10 | 57.6 | |
| Outcome RewardBackbone=Qwen3-4B2025.10 | 56.3 | |
| Ours (Dense)Backbone=Qwen3-4B2025.10 | 53.7 | |
| SFTBackbone=Qwen2.5-7B2025.10 | 53 | |
| Ours (Interval)Backbone=Qwen3-4B2025.10 | 52.1 | |
| Ours (Sparse)Backbone=Qwen2.5-7B2025.10 | 25.2 |