List your company

Multiple Choice Question Answering on GPQA, ARC, MMLU (OOD)

42GPQA AccuracyOverSFT+RL
38.67239.53640.441.264Jun 7, 2026

Evaluation results

MethodLinks
OverSFT+RLBackbone=Qwen3-8B2026.064275.961.759.952.5
Rejuvenation+RLBackbone=Qwen3-8B2026.064178.262.460.553.9
ModSFT+RLBackbone=Qwen3-8B2026.0640.577.962.560.353.8
DFT+RLBackbone=Qwen3-8B2026.0638.876.862.559.351.6