Haotian Sun

Matryoshka Pilot: Learning to Drive Black-Box LLMs with LLMs

Changhao Li
Yuchen Zhuang
Rushi Qiang
Haotian Sun
Hanjun Dai
Chao Zhang
Bo Dai
Accepted at NeurIPS'25,

Abstract

Matryoshka Pilot proposes a framework that uses lightweight LLMs to “drive” black-box LLMs on complex reasoning and agentic tasks. By decomposing control decisions across nested layers of increasing capability, the pilot models progressively refine prompts, tool calls, and plans handed to the frontier black-box model. The method achieves strong improvements in task success and cost efficiency across a suite of reasoning and decision-making benchmarks.

Materials

Project
PDF
DOI

BibTeX

			
@inproceedings{li2025matryoshka,
    title={Matryoshka Pilot: Learning to Drive Black-Box LLMs with LLMs},
    author={Changhao Li and Yuchen Zhuang and Rushi Qiang and Haotian Sun and Hanjun Dai and Chao Zhang and Bo Dai},
    booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
    year={2025}
}