Mathematical Reasoning on GSM8K (test) (Acc, Time)
82.34AccuracyLatent-GRPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Latent-GRPOModel Scale=Qwen3-4B2026.01 | 82.34 | 658.21 | |
| Rule-basedModel Scale=Qwen3-4B2026.01 | 79.87 | 651.45 | |
| Latent-GRPOModel Scale=Qwen3-1.7B2026.01 | 73.88 | 492.34 | |
| LLM-as-JudgeModel Scale=Qwen3-4B2026.01 | 72.12 | 1,411.72 | |
| Rule-basedModel Scale=Qwen3-1.7B2026.01 | 71.55 | 488.63 | |
| LLM-as-JudgeModel Scale=Qwen3-1.7B2026.01 | 64.2 | 1,032.55 | |
| Latent-GRPOModel Scale=Qwen3-0.6B2026.01 | 61.25 | 431.18 | |
| Rule-basedModel Scale=Qwen3-0.6B2026.01 | 58.41 | 434.61 | |
| LLM-as-JudgeModel Scale=Qwen3-0.6B2026.01 | 53.52 | 768.42 |