Mathematical Reasoning on JEE Math
48.5JEE Math Reasoning Score (s)Oracle internal trace
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Oracle internal traceVictim / Teacher=Qwen3-14B, Student supervision=Internal trace2026.05 | 48.5 | 51.2 | |
| Oracle internal traceVictim / Teacher=Qwen3-32B, Student supervision=Internal trace2026.05 | 46.4 | 49.3 | |
| REPCategory=REP exposed trace, Victim / Teacher=Qwen3-32B, Student supervision=Exposed trace, all valid2026.05 | 38.1 | 42.2 | |
| REPCategory=REP exposed trace, Victim / Teacher=Qwen3-32B, Student supervision=Exposed trace, answer-clean2026.05 | 36.4 | 41.1 | |
| REPCategory=REP exposed trace, Victim / Teacher=Qwen3-14B, Student supervision=Exposed trace, answer-clean2026.05 | 35.2 | 39.5 | |
| REPCategory=REP exposed trace, Victim / Teacher=Qwen3-14B, Student supervision=Exposed trace, all valid2026.05 | 33.5 | 38.9 | |
| No distillationVictim / Teacher=–, Student supervision=–2026.05 | 32.2 | 35.9 | |
| Control supervisionVictim / Teacher=Qwen3-14B, Student supervision=Answer only2026.05 | 31.8 | 35.9 | |
| Control supervisionVictim / Teacher=Qwen3-32B, Student supervision=Answer only2026.05 | 31.4 | 35.3 | |
| Control supervisionVictim / Teacher=Qwen3-32B, Student supervision=Summary of reasoning trace2026.05 | 25.8 | 29.4 | |
| Control supervisionVictim / Teacher=Qwen3-14B, Student supervision=Summary of reasoning trace2026.05 | 25.7 | 29.2 | |
| TIAVictim / Teacher=Qwen3-14B, Student supervision=Trace inversion attack2026.05 | 23.9 | 26.3 | |
| TIAVictim / Teacher=Qwen3-32B, Student supervision=Trace inversion attack2026.05 | 19.9 | 21.8 |