The field of reinforcement learning is moving towards more efficient and effective exploration methods. Recent developments have focused on addressing the limitations of existing exploratory bonus methods, which often fail to promote discovery of uncertain regions. New frameworks and techniques have been introduced to counteract divergence-induced bias and unify prior heuristic bonuses. These innovations have led to significant improvements in alignment tasks and mathematical reasoning benchmarks. Notably, the integration of verifier signals with reward-model scores has shown promise in advancing reasoning capabilities. Some noteworthy papers include: General Exploratory Bonus, which provides a principled solution for optimistic exploration, and Token Hidden Reward, which offers a fine-grained mechanism for controlling exploration and exploitation. Additionally, lambda-GRPO and Hybrid Reinforcement have demonstrated the effectiveness of learning token preferences and hybrid reward design in improving reasoning capabilities.