Group Entropy-Controlled Policy Optimization

Benchmark Model Rank Results
instruction-following-on-ifbenchGEPO (Qwen3.5-9B)–Accuracy: 80.7
mathematical-reasoning-on-aime-2025GEPO (Qwen3.5-9B)–Accuracy: 91.4
multimodal-reasoning-on-mathvistaGEPO (Qwen3.5-9B)–Accuracy: 84.9
multimodal-reasoning-on-mmmu-proGEPO (Qwen3.5-9B)–Accuracy: 76.6