Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening

Open paper

Benchmark	Model	Rank	Results
mathematical-reasoning-on-minerva-math	SP3O Qwen3-4B-Base	#13	Accuracy: 47.93
mathematical-reasoning-on-olympiadbench	SP3O Qwen3-8B-Base	#21	Accuracy: 54.5