List your company

Mathematical Reasoning on MGSM High tier

88.9Reward+RL Free
87.54887.89988.2588.601Jun 5, 2026

Evaluation results

MethodLinks
+RL FreeBackbone=Qwen3-4B, RL strategy=Free2026.0688.992
BaselineBackbone=Qwen3-4B, RL strategy=None2026.0688.61
+RL GateBackbone=Qwen3-4B, RL strategy=Gate, Protocol=confidence-gated GSPO2026.0688.621
+RL OTCBackbone=Qwen3-4B, RL strategy=OTC2026.0688.22
+RL FlatBackbone=Qwen3-4B, RL strategy=Flat2026.0687.61