Medical Reasoning on MMLU-Pro (Biology, Hindi)
55.6AccuracyGPT-4o
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4oZero-shot=true2026.05 | 55.6 | 6.3 | |
| HiMed-8BBackbone=LLaMA-3.1-8B, Zero-shot=true2026.05 | 53.8 | 16.1 | |
| HiMed-8B (w/o DSR-RL)Training Stage=without DSR-RL, Zero-shot=true2026.05 | 49.7 | 19.8 | |
| HiMed-8B (w/ LA only)Training Stage=LA only, Zero-shot=true2026.05 | 47 | 21.1 |