Heterogeneous Agent Collaborative Reinforcement Learning

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kQwen3-4B-Instruct (HACPO)–Accuracy: 94.6
math-word-problem-solving-on-mathQwen3-4B-Instruct (HACPO)–Accuracy: 94.3
mathematical-reasoning-on-aime-2025Qwen3-4B-Instruct (HACPO)–Accuracy: 75.7
mathematical-reasoning-on-math-500Qwen3-4B-Instruct (HACPO)–Accuracy: 94.8
mathematical-reasoning-on-minerva-mathQwen3-4B-Instruct (HACPO)–Accuracy: 45.2