InftyThink+: Effective and Efficient Infinite-Horizon Reasoning via Reinforcement Learning

Benchmark Model Rank Results
mathematical-reasoning-on-math-500InftyThink+ (Qwen3-4B-Base, task-only RL)#38Accuracy: 94.09
mathematical-reasoning-on-aime24InftyThink+ (Qwen3-4B-Base, task-only RL)#77Acc: 52.29
multiple-choice-question-answering-mcqa-on-gpqa-diamondInftyThink+ (Qwen3-4B-Base, task-only RL)#116Accuracy: 48.99