[2506.06303] Reward Is Enough: LLMs Are In-Context Reinforcement Learners

Computer Science > Machine Learning arXiv:2506.06303 (cs) [Submitted on 21 May 2025 (v1), last revised 25 Mar 2026 (this version, v5)] Title:Reward Is Enough: LLMs Are In-Context Reinforcement Learners Authors:Kefan Song, Amir Moeini, Peng Wang, Lei Gong, Rohan Chandra, Shangtong Zhang, Yanjun Qi View a PDF of the paper titled Reward Is Enough: LLMs Are In-Context Reinforcement Learners, by Kefan Song and 6 other authors View PDF Abstract:Reinforcement learning (RL) is a framework for solving sequential decision-making problems. In this work, we demonstrate that, surprisingly, RL emerges during the inference time of large language models (LLMs), a phenomenon we term in-context RL (ICRL). To reveal this capability, we introduce a simple multi-round prompting framework, we call ICRL prompting, for inference-time self-improvement. The goal of ICRL prompting is to guide LLMs to perform reinforcement learning during inference for self-improvement on a given task. After each response, the model receives numerical scalar feedback, denoted as a reward. In the next round, we prompt the LLM again together with a context that concatenates all prior responses and their associated rewards. We consistently observe that response quality improves as the context grows. In other words, the LLM can optimize scalar reward signals during inference, exhibiting behavior analogous to reinforcement learning. We evaluate ICRL prompting on Game of 24, creative writing, ScienceWorld, and Olympiad-lev...