Language Understanding on MMLU Low tier
66.7Reward+RL Free
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| +RL FreeBackbone=Qwen3-4B, RL strategy=Free2026.06 | 66.7 | 65 | |
| +RL GateBackbone=Qwen3-4B, RL strategy=Gate, Protocol=confidence-gated GSPO2026.06 | 61.5 | 51 | |
| +RL FlatBackbone=Qwen3-4B, RL strategy=Flat2026.06 | 41.9 | 1 | |
| +RL OTCBackbone=Qwen3-4B, RL strategy=OTC2026.06 | 41.2 | 2 | |
| BaselineBackbone=Qwen3-4B2026.06 | 40.3 | 2 |