On-Policy Self-Distillation for Reasoning Compression
Open paper
Benchmark
Model
Rank
Results
mathematical-reasoning-on-aime24
OPSDC (Qwen3-14B, 30K budget)
#40
Acc: 76.3
mathematical-reasoning-on-math-500
Qwen3-8B (OPSDC)
#65
Accuracy: 86.6
mathematical-reasoning-on-aime-2025
Qwen3-14B (OPSDC)
#66
Accuracy: 61.7
Rank counts only results with a code link.