s3: You Don't Need That Much Data to Train a Search Agent via RL

Retrieval-augmented generation (RAG) systems empower large language models (LLMs) to access external knowledge during inference. Recent advances have enabled LLMs to act as search agents via reinforcement learning (RL), improving information acquisition through multi-turn interactions with retrieval engines. However, existing approaches either optimize retrieval using search-only metrics (e.g., NDCG) that ignore downstream utility or fine-tune the entire LLM to jointly reason and retrieve-entangling retrieval with generation and limiting the real search utility and compatibility with frozen or proprietary models. In this work, we propose s3, a lightweight, modelagnostic framework that decouples the searcher from the generator and trains the searcher using a Gain Beyond RAG reward: the improvement in generation accuracy over naïve RAG. s3 requires only 2.4k training samples to outperform baselines trained on over 70× more data, consistently delivering stronger downstream performance across six general QA and five medical QA benchmarks. 1 Retrieval Generation Retrieval Generation Retrieval Generation distill SFT Active RAG (Zero-Shot) Retrieval Generation Generation Outcome as Signal (e.g., Search-R1) RL Retrieval Generation Retrieval Outcome as Signal (e.g., DeepRetrieval) RL Retrieval RL s3: Gain Beyond RAG as Signal Generation Classic RAG Pre-RLVR