Mathematical Reasoning on MGSM Low tier
86.2Reward+RL Free
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| +RL FreeBackbone=Qwen3-4B, RL strategy=Free2026.06 | 86.2 | 93 | |
| +RL GateBackbone=Qwen3-4B, RL strategy=Gate, Protocol=confidence-gated GSPO2026.06 | 83.2 | 57 | |
| +RL FlatBackbone=Qwen3-4B, RL strategy=Flat2026.06 | 65.4 | 14 | |
| +RL OTCBackbone=Qwen3-4B, RL strategy=OTC2026.06 | 61.6 | 17 | |
| BaselineBackbone=Qwen3-4B2026.06 | 51.4 | 4 |