Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Multi-Agent Collaboration via Evolving Orchestration
Authors: Yufan Dang, Chen Qian, Xueheng Luo, Jingru Fan, Zihao Xie, Ruijie Shi, Weize Chen, Cheng Yang, Xiaoyin Che, Ye Tian, Xuantang Xiong, Lei Han, Zhiyuan Liu, Maosong Sun
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on closed- and open-domain scenarios show that this method achieves superior performance with reduced computational costs. Empirical results on both closed- and open-domain scenarios demonstrate that our approach consistently yields more effective solutions while requiring less computational overhead. |
| Researcher Affiliation | Collaboration | Tsinghua University Shanghai Jiao Tong University Beijing University of Posts and Telecommunications Siemens Tencent Robotics X |
| Pseudocode | No | The paper describes the methodology using mathematical formulations and descriptive text, but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github.com/Open BMB/Chat Dev/tree/puppeteer. |
| Open Datasets | Yes | GSMHard[18] features arithmetic problems... MMLU-Pro[67] is a comprehensive benchmark... SRDD[44] consists of real-world textual software requirements... Common Gen-Hard[39] challenges agents to generate coherent sentences... |
| Dataset Splits | No | The paper uses datasets such as GSMHard[18], MMLU-Pro[67], SRDD[44], and Common Gen-Hard[39] for evaluation, but it does not explicitly provide specific details on how these datasets were split into training, validation, or test sets. |
| Hardware Specification | Yes | GPU NVIDIA A800, 8 GPUs used |
| Software Dependencies | No | The paper mentions implementing agents with Python capabilities and using Llama-3.1 as a policy, but it does not specify version numbers for any key software libraries or frameworks used in the experiments. |
| Experiment Setup | Yes | The policy is initialized with a variant of Llama-3.14, using default settings: episode length to 4, parallel exploration up to 3, λ = 0.1, and γ = 0.99. |