WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents
This work addresses the problem of limited web browsing capabilities in LLM-based agents for complex, multi-step information retrieval tasks, representing a strong specific gain in the domain of web agents.
The authors tackled the challenge of training long-horizon web agents by addressing the scarcity of challenging data for information seeking, introducing WebExplorer, a data generation method that creates complex query-answer pairs. Their resulting 8B-sized model achieved state-of-the-art performance on benchmarks, outperforming larger models like WebSailor-72B and achieving high accuracy with up to 100 tool calling turns.
The paradigm of Large Language Models (LLMs) has increasingly shifted toward agentic applications, where web browsing capabilities are fundamental for retrieving information from diverse online sources. However, existing open-source web agents either demonstrate limited information-seeking abilities on complex tasks or lack transparent implementations. In this work, we identify that the key challenge lies in the scarcity of challenging data for information seeking. To address this limitation, we introduce WebExplorer: a systematic data generation approach using model-based exploration and iterative, long-to-short query evolution. This method creates challenging query-answer pairs that require multi-step reasoning and complex web navigation. By leveraging our curated high-quality dataset, we successfully develop advanced web agent WebExplorer-8B through supervised fine-tuning followed by reinforcement learning. Our model supports 128K context length and up to 100 tool calling turns, enabling long-horizon problem solving. Across diverse information-seeking benchmarks, WebExplorer-8B achieves the state-of-the-art performance at its scale. Notably, as an 8B-sized model, WebExplorer-8B is able to effectively search over an average of 16 turns after RL training, achieving higher accuracy than WebSailor-72B on BrowseComp-en/zh and attaining the best performance among models up to 100B parameters on WebWalkerQA and FRAMES. Beyond these information-seeking tasks, our model also achieves strong generalization on the HLE benchmark even though it is only trained on knowledge-intensive QA data. These results highlight our approach as a practical path toward long-horizon web agents.