Agentic Aggregation for Parallel Scaling of Long-Horizon Agentic Tasks

TL;DR

This paper introduces AggAgent, a novel aggregation method that treats parallel trajectories as an environment, enabling effective synthesis of long-horizon agentic task outputs with minimal overhead.

Contribution

The paper presents AggAgent, a new aggregation framework that improves parallel scaling of long-horizon agentic tasks by treating trajectories as an environment for better synthesis.

Findings

AggAgent outperforms existing methods by up to 5.3% on benchmarks.

It achieves 10.3% improvement on deep research tasks.

Aggregation cost remains bounded by a single agentic rollout.

Abstract

We study parallel test-time scaling for long-horizon agentic tasks such as agentic search and deep research, where multiple rollouts are generated in parallel and aggregated into a final response. While such scaling has proven effective for chain-of-thought reasoning, agentic tasks pose unique challenges: trajectories are long, multi-turn, and tool-augmented, and outputs are often open-ended. Aggregating only final answers discards rich information from trajectories, while concatenating all trajectories exceeds the model's context window. To address this, we propose AggAgent, an aggregation agent that treats parallel trajectories as an environment. We equip it with lightweight tools to inspect candidate solutions and search across trajectories, enabling it to navigate and synthesize information on demand. Across six benchmarks and three model families (GLM-4.7, Qwen3.5, MiniMax-M2.5), AggAgent outperforms all existing aggregation methods-by up to 5.3% absolute on average and 10.3% on two deep research tasks-while adding minimal overhead, as the aggregation cost remains bounded by a single agentic rollout. Our findings establish agentic aggregation as an effective and cost-efficient approach to parallel test-time scaling.