Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search

TL;DR

Long-horizon agentic search faces context-window and tool-budget bottlenecks that degrade performance and inflate costs. The authors introduce Slim, a lightweight framework with three modular tools—search, browse, and summarization—that keeps context concise while enabling longer, more targeted searches; summarization occurs every $n$ turns to control noise. Empirically, Slim with base model $o_3$ outperforms open-source baselines on BrowseComp and Humanity's Last Exam by 8 and 4 points respectively, at 4–6x fewer tool calls and lower costs, and scales to around 150 turns. Additionally, the paper provides an automated trajectory-level analysis pipeline and an error taxonomy, showing Slim reduces hallucinations and clarifying failure modes, thereby offering a practical blueprint for building effective long-horizon search agents.

Abstract

Long-horizon agentic search requires iteratively exploring the web over long trajectories and synthesizing information across many sources, and is the foundation for enabling powerful applications like deep research systems. In this work, we show that popular agentic search frameworks struggle to scale to long trajectories primarily due to context limitations-they accumulate long, noisy content, hit context window and tool budgets, or stop early. Then, we introduce SLIM (Simple Lightweight Information Management), a simple framework that separates retrieval into distinct search and browse tools, and periodically summarizes the trajectory, keeping context concise while enabling longer, more focused searches. On long-horizon tasks, SLIM achieves comparable performance at substantially lower cost and with far fewer tool calls than strong open-source baselines across multiple base models. Specifically, with o3 as the base model, SLIM achieves 56% on BrowseComp and 31% on HLE, outperforming all open-source frameworks by 8 and 4 absolute points, respectively, while incurring 4-6x fewer tool calls. Finally, we release an automated fine-grained trajectory analysis pipeline and error taxonomy for characterizing long-horizon agentic search frameworks; SLIM exhibits fewer hallucinations than prior systems. We hope our analysis framework and simple tool design inform future long-horizon agents.

Figures (8)

• Figure 1: With o3 as the base model, Slim achieves better performance than existing frameworks on both BrowseComp and HLE while using more than 4-6x fewer tool calls and lower overall costs, which account for LLM token usage and tool costs.
• Figure 2: Example queries and their relevant documents for BrowseComp wei2025browsecompsimplechallengingbenchmark and HLE phan2025humanitysexam.
• Figure 3: Each framework exhibits distinct outcome trends---ReAct predominantly runs out of context window, while Search-o1 is often limited by the tool budget (T). We exclude GPT-R due to its predefined workflow---the outcome can only be either correct or incorrect.
• Figure 4: Compared to ReAct and Search-o1, the cooperation between search, browse, and summarization modules allows Slim to accumulate shorter contexts and less noisy information after exploring the same amount of searches.
• Figure 5: With o4-mini as the base model, Slim consistently outperforms other baselines on BrowseComp while using fewer tool calls and lower overall costs. On HLE, Slim can achieve overall higher performance and use fewer tool calls.