ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents

TL;DR

ARC introduces an active, reflection-driven framework for managing context in long-horizon information-seeking agents, significantly improving their performance by dynamically maintaining coherent internal states over extended reasoning tasks.

Contribution

This work is the first to formulate context management as an active, reflection-driven process, moving beyond static summarization to improve long-term reasoning in language models.

Findings

ARC outperforms passive methods by up to 11% accuracy

Active context management reduces degradation in long-horizon tasks

Demonstrates effectiveness on challenging benchmarks

Abstract

Large language models are increasingly deployed as research agents for deep search and long-horizon information seeking, yet their performance often degrades as interaction histories grow. This degradation, known as context rot, reflects a failure to maintain coherent and task-relevant internal states over extended reasoning horizons. Existing approaches primarily manage context through raw accumulation or passive summarization, treating it as a static artifact and allowing early errors or misplaced emphasis to persist. Motivated by this perspective, we propose ARC, which is the first framework to systematically formulate context management as an active, reflection-driven process that treats context as a dynamic internal reasoning state during execution. ARC operationalizes this view through reflection-driven monitoring and revision, allowing agents to actively reorganize their working context when misalignment or degradation is detected. Experiments on challenging long-horizon information-seeking benchmarks show that ARC consistently outperforms passive context compression methods, achieving up to an 11% absolute improvement in accuracy on BrowseComp-ZH with Qwen2.5-32B-Instruct.