Rubric Entailment on RiceChem (test)
86.8AccuracyRoBERTa+MNLI
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RoBERTa+MNLItraining=fine-tuned2026.02 | 86.8 | — | |
| Autorubricjudge=Gemini-3-Flash, shots=5-shot2026.02 | 80 | 0.92 | |
| Autorubricjudge=Gemini-3-Flash, shots=3-shot2026.02 | 79 | 0.77 | |
| Autorubricjudge=Gemini-3-Flash, shots=0-shot2026.02 | 77.2 | 0.51 | |
| GPT-4shots=0-shot2026.02 | 70.9 | — |