AFM-WebAgent-7B-sft

Model Introduction

We introduce Agent Foundation Models (AFMs), a new family built on Qwen2.5 that natively perform end-to-end, multi-turn, multi-tool problem solving—without external frameworks or manual prompting. Built on the Chain-of-Agents (CoA) paradigm, each AFM dynamically activates specialized tool and role-playing agents inside a single forward pass, emulating the cooperative reasoning of a full multi-agent system. To train these models, we distilled high-performing multi-agent trajectories into agentic supervised-fine-tuning data and further optimized performance with agentic reinforcement learning on verifiable tasks. AFMs set new state-of-the-art results on benchmarks for both web and code agents, and we release all model weights, training code, and datasets to accelerate future research on agentic AI. For more details, please refer to our Projects, paper and GitHub.

Model Downloads

Data Downloads

• AFM-CodeAgent-SFT-Dataset
• AFM-CodeAgent-RL-Dataset
• AFM-WebAgent-SFT-Dataset
• AFM-WebAgent-RL-Dataset
• AFM-MHQA-SFT-Dataset
• AFM-MHQA-RL-Dataset

Citation

If you find AFM useful in your research or applications, we would appreciate it if you could cite our work:

@misc{li2025chainofagentsendtoendagentfoundation,
      title={Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL}, 
      author={Weizhen Li and Jianbo Lin and Zhuosong Jiang and Jingyi Cao and Xinpeng Liu and Jiayu Zhang and Zhenqiang Huang and Qianben Chen and Weichen Sun and Qiexiang Wang and Hongxuan Lu and Tianrui Qin and Chenghao Zhu and Yi Yao and Shuying Fan and Xiaowan Li and Tiannan Wang and Pai Liu and King Zhu and He Zhu and Dingfeng Shi and Piaohong Wang and Yeyi Guan and Xiangru Tang and Minghao Liu and Yuchen Eleanor Jiang and Jian Yang and Jiaheng Liu and Ge Zhang and Wangchunshu Zhou},
      year={2025},
      eprint={2508.13167},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2508.13167}, 
}