Predicting Privileged Information Effectiveness for On-Policy Self-Distillation

Open paper
Benchmark Model Rank Results
agents-on-alfworldPI-OPSD (Qwen3-4B-Instruct-2507)#2Success Rate: 95.13
agents-on-bench-retailPI-OPSD (Qwen3-4B-Instruct-2507)#9Accuracy: 55.04

Rank counts only results with a code link.