The field of constraint programming and reinforcement learning is moving towards more efficient and effective methods for solving complex problems. Researchers are exploring new approaches to automate the translation of natural language problem descriptions into formal constraint models, and to improve the reasoning and code generation capabilities of multi-agent systems. One notable direction is the use of agentic strategies and decoupled role design to reduce cognitive load interference and promote stable reasoning-coding coordination. Another area of focus is the development of novel reward mechanisms, such as verifiable stepwise rewards and gated reward accumulation, to address the challenges of reward sparsity and misalignment in long-horizon reinforcement learning tasks. These advances have the potential to significantly improve the performance and efficiency of constraint programming and reinforcement learning systems. Noteworthy papers include: CP-Agent, which presents a new approach to automating constraint programming using a pure agentic strategy, and Reducing Cognitive Load in Multi-Agent Reinforcement Learning, which proposes a dual-agent hybrid framework to decouple reasoning and code generation. Promoting Efficient Reasoning with Verifiable Stepwise Reward is also notable for its novel rule-based verifiable stepwise reward mechanism, and Stabilizing Long-term Multi-turn Reinforcement Learning with Gated Rewards for its introduction of Gated Reward Accumulation. Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models is also worth mentioning for its exploration of Pass@k as a reward metric to train policy models.