Not Worth Another Token: Marginal Value Estimation for Efficient Deep Research Agents

2026-08-08

AI Summary

This work addresses the challenge of context inflation in long-horizon research agents, where iterative retrieval and synthesis often suffer from rapidly diminishing marginal utility of new evidence, leading to high computational costs, latency, and noisy outputs. The authors propose a multi-stage context pruning framework that integrates lightweight heuristic criteria and learned value models during pre-retrieval, post-retrieval, and pre-synthesis phases. Through the first systematic comparison of diverse pruning strategies, they demonstrate that the timing of pruning exerts a far greater impact on efficiency than the choice of scoring rule. Experiments show that lightweight heuristics can reduce token usage by up to 73% with minimal quality degradation, while learned methods excel under specific trade-offs. No single strategy dominates across all dimensions—quality, efficiency, and faithfulness—offering practical guidance for designing efficient long-horizon agents.

Abstract

Long-horizon research agents solve open-ended tasks through iterative retrieval, aggregation, and synthesis, but context grows rapidly while the marginal value of additional evidence often declines. This leads to unnecessary token cost, higher latency, and noisier inputs for final report generation. We study marginal value estimation for context management in deep research agents and present the first systematic stage-aware comparison of pruning strategies across the pipeline. We evaluate lightweight heuristic criteria and a learned value model at pre-retrieval, post-retrieval, and pre-synthesis stages. Our results show that pruning effectiveness depends more on where pruning is applied than on the specific scoring rule: early pruning yields the largest end-to-end savings, while later pruning mainly refines the final synthesis context. Lightweight heuristics reduce token usage by up to 73% with little quality degradation, learned pruning remains competitive on selected trade-offs, and no single method dominates across quality, efficiency, and faithfulness. These findings provide practical guidance for designing efficient long-horizon agentic systems.

Problem

• marginal value
• context management
• deep research agents
• token efficiency
• long-horizon reasoning

Innovation

• marginal value estimation
• context pruning
• deep research agents
• token efficiency
• stage-aware optimization