AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation

Siyu Wang; Ruotian Lu; Zhihao Yang; Yuchao Wang; Yanzhou Zhang; Lei Xu; Qimin Xu; Guojun Yin

cs.MAFeb 19, 2026

AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation

Authors: Siyu Wang, Ruotian Lu, Zhihao Yang, Yuchao Wang, Yanzhou Zhang, Lei Xu, Qimin Xu, Guojun Yin,

Organizations: Shanghai Jiao Tong University · Meituan

Abstract

Large language model(LLM)-driven multi-agent systems(MAS) coordinate specialized agents through predefined interaction topologies and have shown promise for complex tasks such as competition-level code generation. Recent studies demonstrate that carefully designed multi-agent workflows and communication graphs can significantly improve code generation performance by leveraging collaborative reasoning. However, existing methods neither adapt topology density to task difficulty nor iteratively refine the topology within an instance using execution feedback, which leads to redundant communication and performance bottlenecks. To address these issues, we propose AgentConductor: a reinforcement learning-optimized MAS with an LLM-based orchestrator agent as its core, which enables end-to-end feedback-driven dynamic generation of interaction topologies. For each query, AgentConductor infers agent roles and task difficulty, then constructs a task-adapted, density-aware layered directed acyclic graph (DAG) topology, underpinned by two key innovations. First, we design a novel topological density function that captures communication-aware mathematical characterizations of multi-agent interactions. Second, we adopt difficulty interval partitioning to avoid excessive pruning for precise topological density upper bound measurement per difficulty level and finer-grained control. Empirically, across three competition-level and two foundational code datasets, AgentConductor achieves state-of-the-art accuracy, outperforming the strongest baseline by up to 14.6% in pass@1 accuracy, 13% in density reduction, and 68% in token cost reduction.

Figures & tables

Figure 1: YAML representation of the topology, its mapping to the actual graph, and the two-turn graph evolution.

Figure 2: Comparison of Topology Structures and Optimization Paradigms Between Our Method and Classic Baselines

Figure 3: Overall framework of the proposed AgentConductor . The approach proceeds in three stages: (1) SFT on diverse topologies to instill structural priors in the base LLM (Qwen-2.5-Instruct-3B); (2) RL with GRPO to learn task-adaptive, difficulty-aware topology policies from execution feedback, yielding the orchestrator agent; and (3) multi-turn dynamic topology generation for end-to-end code problem solving.

Panel of Table 1 — Table 1: Rewards for Topology Validation and Code Execution Errors

Method	Contest-level Code Generation				Basic Code Generation			Avg.
	APPs	LiveCodeBench	CodeContests	Avg.	HumanEval	MBPP	Avg.	Avg.
Vanilla
GPT-4o-mini	$20.3_{\text{(\textpm 0.2)}}$	$26.3_{\text{(\textpm 0.2)}}$	$18.6_{\text{(\textpm 0.4)}}$	$21.7_{\text{(\textpm 0.3)}}$	$87.6_{\text{(\textpm 0.2)}}$	$73.5_{\text{(\textpm 0.1)}}$	$80.5_{\text{(\textpm 0.1)}}$	$51.1_{\text{(\textpm 0.2)}}$
Classical Multi-Agent Systems (No Workflow/Topology Optimization)
AutoGen	$23.6_{\text{(\textpm 2.3)}}$	$30.2_{\text{(\textpm 1.5)}}$	$20.8_{\text{(\textpm 1.9)}}$	$24.9_{\text{(\textpm 1.9)}}$	$90.4_{\text{(\textpm 0.8)}}$	$92.3_{\text{(\textpm 0.4)}}$	$91.4_{\text{(\textpm 0.6)}}$	$58.1_{\text{(\textpm 1.3)}}$
MetaGPT	$\underline{51.3}_{\text{(\textpm 1.4)}}$	$42.8_{\text{(\textpm 1.3)}}$	$35.6_{\text{(\textpm 1.2)}}$	$\underline{43.2}_{\text{(\textpm 1.3)}}$	$95.8_{\text{(\textpm 0.2)}}$	$92.3_{\text{(\textpm 0.3)}}$	$94.1_{\text{(\textpm 0.2)}}$	$\underline{68.7}_{\text{(\textpm 0.6)}}$

Table 2: Main performance of AgentConductor on three competition-level and two basic code generation datasets (mean ± std over 3 runs).

Dataset	Method	Perf.	Prompt	Comp.	$\mathcal{S}_{\text{complex}}$ (↑)
APPS	AFlow	35.4	531450	184800	3.7
	FlowReasoner	39.1	437250	148050	2.4
	Chain-of-Agents (32B)	41.6	334650	134250	4.1
	GPTSwarm	36.5	381450	155400	3.5
	AgentPrune (Layered)	39.3	364950	141150	3.8
	MacNet (Layered)	36.9	472950	200100	2.9

Table 3: APPS results comparing AgentConductor with baselines on performance, cost, and average topology density.

Figure 4: (a) APPS results showing performance, average graph density ( $\mathcal{S}_{\text{complex}}$ ↑ sparser), and completion tokens, with circle size indicating token savings (diameter↑ more). (b) Code generation performance comparison of representative baselines.

Group	Method	APPS			HumanEval
		Perf.	$\mathcal{S}_{\text{complex}}$ (↑)	Valid (%)	Perf.	$\mathcal{S}_{\text{complex}}$ (↑)	Valid (%)
	Full Model	58.8	5.2	100	97.5	5.8	100
Training Strategies	w/o SFT	–	–	15	–	–	13
Training Strategies	w/o RL	29.8	2.7	56.5	90.2	3.2	57.2
Reward	w/o $r_{e}(\mathcal{E}_{\text{yaml\_errors}})$	30.3	2.9	56.8	91.4	3.0	58.1
	w/o $r_{e}(\mathcal{E}_{\text{code\_errors}})$	35.5	5.0	96.4	93.1	5.6	99.2

Table 4: Ablation study on Training Strategies and Reward Design.

Figure 5: Comparison of the average topology density (

\mathcal{S}_{\text{complex}}

↑ sparser) across three competition-level code datasets at three difficulty levels.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: The figure shows the system prompt for the orchestrator agent.

Figure 7: The figure shows the dynamics of three key metrics during RL training: (a) training reward, (b) average number of valid two-turn topologies, and (c) validation reward. The results indicate that our method progressively converges toward generating topologies with reasonable density and achieving accurate code problem solving in later training stages.

Figure 8: The figure shows the generated interaction topologies for two problem cases at each difficulty level.

Method	Backbone	GAIA L1	GAIA L2	GAIA L3	GAIA Avg.	HLE Avg.	PopQA
Chain-of-Agents	7B	$69.2_{\text{(\textpm 0.8)}}$	$50.9_{\text{(\textpm 0.7)}}$	$33.3_{\text{(\textpm 1.1)}}$	$50.8_{\text{(\textpm 0.8)}}$	$18.0_{\text{(\textpm 0.6)}}$	$46.5_{\text{(\textpm 1.3)}}$
AgentConductor (ours)	3B	$72.0_{\text{(\textpm 0.4)}}$	$53.4_{\text{(\textpm 0.3)}}$	$36.1_{\text{(\textpm 0.5)}}$	$53.8_{\text{(\textpm 0.4)}}$	$22.6_{\text{(\textpm 0.2)}}$	$50.3_{\text{(\textpm 0.3)}}$

Appendix

Table 5: Cross-domain evaluation of AgentConductor on GAIA, HLE, and PopQA. Results are reported as mean

\pm

std over three seeds.

Panel of Figure 14 — Algorithm 1 Online Topology Generation Workflow of AgentConductor

LLM Agent Workflow Optimization

LLM agent workflow optimization studies how to design and run multi-step agents for better task success, reliability, latency, and cost.

See all 243 papers

Learning How Much to Collaborate: Difficulty-Aware Topology Selection for Multi-Agent Code Generation
Sep 12, 2026Yunsong HongMulti-Agent Coding Agentic Code Generation
School of Computer Science, The University of Sydney, Sydney, NSW 2006, Australia
Retrieval-Conditioned Topology Selection with Provable Budget Conservation for Multi-Agent Code Generation
May 7, 2026Abhijit Talluri, Pujith Anne, Bhagavan Choudary Pendiyala +1Multi-Agent Orchestration Multi-Agent Coding
Independent Researcher
GoAgent: Group-of-Agents Communication Topology Generation for LLM-based Multi-Agent Systems
Mar 20, 2026Hongjiang Chen, Xin Zheng, Yixin Liu +7Multi-Agent Communication Topology Multi-Agent Communication
School of Cyberspace, Hangzhou Dianzi University, Hangzhou, China · School of Computing Technologies, RMIT University, Melbourne, Australia · School of Information and Communication Technology, Griffith University, Goldcoast, Australia

Sep 12, 2026J/K move · Enter open · S save

Learning How Much to Collaborate: Difficulty-Aware Topology Selection for Multi-Agent Code Generation

Yunsong Hong

School of Computer Science, The University of Sydney, Sydney, NSW 2006, Australia

Multi-Agent Coding Agentic Code Generation Multi-Agent Communication Topology Code Generation Evaluation

Open paper AskPDF arXiv

Multi-agent systems for code generation are deployed with a single communication topology, chosen once for every problem. This is the wrong granularity. Evaluating five topologies on 614 problems from APPS, HumanEval+ and LiveCodeBench, we find that the advantage of hierarchical collaboration over a single agent grows from 2.4 points of pass@1 on the easiest third of problems to 21.1 points on the hardest third, while its token cost stays about ten times higher. We propose the Difficulty-Aware Topology Selector (DATS), which predicts each topology's probability of solving a problem and selects the one maximising predicted success minus cost. Its predictor is a graph network that treats the five topologies as nodes of a connectivity order rather than independent labels, worth 1.7 points over a flat multi-label head. Because the cost penalty is a single scalar recalibrable without retraining, routers compare at equal spend: under this budget-matched protocol six cost-aware methods span 21.6 percentage points, and two baselines leading DATS fall behind once calibrated to it. Fixed at 40% of the always-hierarchical cost, DATS reaches 77.7% pass@1 against 73.6% (always-hierarchical) and 74.3% (strongest learned competitor), all eleven pairwise McNemar comparisons surviving Holm-Bonferroni correction. The 4.1-point gain holds across four backbones spanning fourteen points of capability, and replacing the 39 interpretable features with a graph network or a pretrained encoder shifts accuracy by at most 1.3 points, never significantly. A cross-domain study on 400 mathematical reasoning problems reproduces the effect, the gap widening from 2.5 to 20.9 points.

Abstract

LLM Agent Workflow Optimization

Explore similar work

Learning How Much to Collaborate: Difficulty-Aware Topology Selection for Multi-Agent Code Generation

Retrieval-Conditioned Topology Selection with Provable Budget Conservation for Multi-Agent Code Generation

GoAgent: Group-of-Agents Communication Topology Generation for LLM-based Multi-Agent Systems