ResearchBenchmarksMathematical Reasoning on GSM8K DE (test)Mathematical Reasoning on GSM8K DE (test)Follow66AccuracyKL-DRTO59.7661.386364.62Mar 27, 2026Evaluation resultsMethodMethodLinksAccuracyKL-DRTO2026.0366χ2-DRTO2026.0365.6PPO2026.0362DPO2026.0362SFT2026.0361.6RTO2026.0360.8GRPO2026.0360