Journal tier classification on Management (held-out)
61.7Accuracypooled Qwen3-30B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| pooled Qwen3-30BN=120, Role=Pooled SFT2026.03 | 61.7 | 61.5 | |
| single-field SFT ensembleN=120, Role=SFT ensemble2026.03 | 60.8 | — | |
| pooled GPT-4.1-nanoN=120, Role=Pooled SFT2026.03 | 52.5 | 48.9 |