Long-horizon research agents tackle open-ended tasks through iterative retrieval, aggregation, and synthesis, but their context grows rapidly while the marginal value of additional evidence often declines. This leads to unnecessary token costs, higher latency, and noisier inputs for final report generation.
In a new paper, researchers from Hugging Face present the first systematic stage-aware comparison of pruning strategies across the pipeline. They evaluate lightweight heuristic criteria and a learned value model at pre-retrieval, post-retrieval, and pre-synthesis stages.
Key findings show that pruning effectiveness depends more on where it 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, while learned pruning remains competitive on selected trade-offs.
The authors note that no single method dominates across quality, efficiency, and faithfulness, providing practical guidance for designing efficient long-horizon agentic systems.