Mathematical Reasoning on Arithmetic Dataset (test)
930Avg Generated TokensSELFCEST
Evaluation results
| Method | Links | |
|---|---|---|
| SELFCESTModel variant=Final2026.02 | 930 | |
| Qwen3-4B low token RLModel variant=low token RL2026.02 | 1,455 | |
| Base Qwen3-4BModel variant=Base2026.02 | 1,924 | |
| Qwen3-4B-ThinkingModel variant=Thinking2026.02 | 6,220 |