@inproceedings{chen-etal-2026-multi-agent-evolve,
title = "Multi-Agent Evolve: {LLM} Self-Improve through Co-evolution",
author = "Chen, Yixing and
Wang, Yiding and
Zhu, Siqi and
Xu, Weixian and
Yu, Haofei and
Feng, Tao and
Zhang, Muhan and
Patwary, Mostofa and
You, Jiaxuan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1734/",
pages = "39563--39593",
ISBN = "979-8-89176-459-0",
abstract = "Reinforcement Learning (RL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets, which pose a burden on scalability and sustainability. Inspired by the success of the paradigm in games and Go, Self-Play methods are deployed to enhance LLM capabilities without human-annotated data. However, these methods primarily depend on a grounded environment for verifiable feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose **Multi-Agent Evolve (MAE)**, a framework in which LLM can self-evolve across diverse tasks from mathematics to general knowledge. MAE implements asymmetric Self-Play by leveraging a triplet of interacting roles (*Proposer, Solver, Judge*) that are instantiated from a single LLM. *Proposer* generates challenging questions, *Solver* optimizes solutions, and *Judge* provides reward signals. These roles are jointly optimized via reinforcement learning, creating a closed-loop system where the model continuously challenges and improves itself. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.86{\%} across multiple benchmarks, significantly outperforming baselines. These results highlight MAE as a scalable, data-efficient method for enhancing LLMs' general reasoning abilities with minimal reliance on human supervision."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-multi-agent-evolve">
<titleInfo>
<title>Multi-Agent Evolve: LLM Self-Improve through Co-evolution</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yixing</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiding</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siqi</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weixian</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haofei</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tao</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Muhan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mostofa</namePart>
<namePart type="family">Patwary</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxuan</namePart>
<namePart type="family">You</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Reinforcement Learning (RL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets, which pose a burden on scalability and sustainability. Inspired by the success of the paradigm in games and Go, Self-Play methods are deployed to enhance LLM capabilities without human-annotated data. However, these methods primarily depend on a grounded environment for verifiable feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose **Multi-Agent Evolve (MAE)**, a framework in which LLM can self-evolve across diverse tasks from mathematics to general knowledge. MAE implements asymmetric Self-Play by leveraging a triplet of interacting roles (*Proposer, Solver, Judge*) that are instantiated from a single LLM. *Proposer* generates challenging questions, *Solver* optimizes solutions, and *Judge* provides reward signals. These roles are jointly optimized via reinforcement learning, creating a closed-loop system where the model continuously challenges and improves itself. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.86% across multiple benchmarks, significantly outperforming baselines. These results highlight MAE as a scalable, data-efficient method for enhancing LLMs’ general reasoning abilities with minimal reliance on human supervision.</abstract>
<identifier type="citekey">chen-etal-2026-multi-agent-evolve</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1734/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>39563</start>
<end>39593</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Multi-Agent Evolve: LLM Self-Improve through Co-evolution
%A Chen, Yixing
%A Wang, Yiding
%A Zhu, Siqi
%A Xu, Weixian
%A Yu, Haofei
%A Feng, Tao
%A Zhang, Muhan
%A Patwary, Mostofa
%A You, Jiaxuan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F chen-etal-2026-multi-agent-evolve
%X Reinforcement Learning (RL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets, which pose a burden on scalability and sustainability. Inspired by the success of the paradigm in games and Go, Self-Play methods are deployed to enhance LLM capabilities without human-annotated data. However, these methods primarily depend on a grounded environment for verifiable feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose **Multi-Agent Evolve (MAE)**, a framework in which LLM can self-evolve across diverse tasks from mathematics to general knowledge. MAE implements asymmetric Self-Play by leveraging a triplet of interacting roles (*Proposer, Solver, Judge*) that are instantiated from a single LLM. *Proposer* generates challenging questions, *Solver* optimizes solutions, and *Judge* provides reward signals. These roles are jointly optimized via reinforcement learning, creating a closed-loop system where the model continuously challenges and improves itself. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.86% across multiple benchmarks, significantly outperforming baselines. These results highlight MAE as a scalable, data-efficient method for enhancing LLMs’ general reasoning abilities with minimal reliance on human supervision.
%U https://aclanthology.org/2026.emnlp-main.1734/
%P 39563-39593
Markdown (Informal)
[Multi-Agent Evolve: LLM Self-Improve through Co-evolution](https://aclanthology.org/2026.emnlp-main.1734/) (Chen et al., EMNLP 2026)
ACL
- Yixing Chen, Yiding Wang, Siqi Zhu, Weixian Xu, Haofei Yu, Tao Feng, Muhan Zhang, Mostofa Patwary, and Jiaxuan You. 2026. Multi-Agent Evolve: LLM Self-Improve through Co-evolution. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 39563–39593, Budapest, Hungary. Association for Computational Linguistics.