List your company

Mathematical Reasoning on MGSM Low tier

86.2Reward+RL Free
50.00859.40468.878.196Jun 5, 2026

Evaluation results

MethodLinks
+RL FreeBackbone=Qwen3-4B, RL strategy=Free2026.0686.293
+RL GateBackbone=Qwen3-4B, RL strategy=Gate, Protocol=confidence-gated GSPO2026.0683.257
+RL FlatBackbone=Qwen3-4B, RL strategy=Flat2026.0665.414
+RL OTCBackbone=Qwen3-4B, RL strategy=OTC2026.0661.617
BaselineBackbone=Qwen3-4B2026.0651.44