Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Small Model as Master Orchestrator: Learning Unified Agent-Tool Orchestration with Parallel Subtask Decomposition

About

Multi-agent systems (MAS) demonstrate clear advantages in tackling complex problems by coordinating diverse agents and external tools. However, most existing orchestration methods rely on static workflows or serial agent scheduling, and are further constrained by heterogeneous interface protocols between tools and agents. This leads to high system complexity and poor extensibility. To mitigate these issues, we propose Agent-as-Tool, a unified parallel orchestration paradigm that abstracts both agents and tools into a standardized, learnable action space with protocol normalization and explicit state feedback. Building on this paradigm, we train a lightweight orchestrator, ParaManager, which decouples planning decisions from subtask solving, enabling state-aware parallel subtask decomposition, delegation, and asynchronous execution. For training, we adopt a two-stage ParaManager training pipeline. It improves robustness by incorporating supervised fine-tuning (SFT) trajectories equipped with recovery mechanisms, and further applies reinforcement learning (RL) to achieve an optimal balance among task success, protocol compliance, diversity, and reasoning efficiency. Experiments show that ParaManager achieves strong performance across multiple benchmarks and exhibits robust generalization under unseen model pools.

Wenzhen Yuan, Wutao Xiong, Fanchen Yu, Shengji Tang, Ting Liu, Tao Chen, Peng Ye, Yuzhuo Fu, Wanli Ouyang, Lei Bai• 2026

Related benchmarks

TaskDatasetResultRank
Mathematical ReasoningAMC
Accuracy (%)95.18
368
Mathematical ReasoningAIME 24
Accuracy86.67
318
Mathematical ReasoningHMMT25
Accuracy (%)63.33
115
Mathematical Problem SolvingAIME 2024
Accuracy86.67
113
General Knowledge ReasoningMMLU-Pro
Accuracy81.43
64
Code GenerationLCB v6
Accuracy42.5
49
Mathematical Problem SolvingAIME 2025
Top-1 Accuracy (%)90
46
Code GenerationLCB v5
Accuracy39.45
45
Mathematical ReasoningAIME25
Accuracy84.17
25
Scientific ReasoningGPQA
Accuracy (%)72.1
25
Showing 10 of 14 rows

Other info

Follow for update