Downstream Accuracy on JEEBench
44.8AccuracyVictim-Trace (oracle)
Evaluation Results
| Method | Links | |
|---|---|---|
| Victim-Trace (oracle)Surrogate=-, Student Model=Qwen2026.03 | 44.8 | |
| Synthesized-TraceSurrogate=R1-Distill, Student Model=Qwen2026.03 | 36.3 | |
| No finetuningSurrogate=-, Student Model=Qwen2026.03 | 28.3 | |
| Victim-Trace (oracle)Surrogate=-, Student Model=Llama2026.03 | 28.1 | |
| Synthesized-TraceSurrogate=R1-Distill, Student Model=Llama2026.03 | 24.2 | |
| Answer+SummarySurrogate=-, Student Model=Qwen2026.03 | 24 | |
| Answer-onlySurrogate=-, Student Model=Qwen2026.03 | 21.6 | |
| Surrogate-TraceSurrogate=R1-Distill, Student Model=Qwen2026.03 | 19.7 | |
| Answer+SummarySurrogate=-, Student Model=Llama2026.03 | 18.2 | |
| Answer-onlySurrogate=-, Student Model=Llama2026.03 | 17.2 | |
| No finetuningSurrogate=-, Student Model=Llama2026.03 | 16.6 | |
| Surrogate-TraceSurrogate=R1-Distill, Student Model=Llama2026.03 | 11.3 |