@inproceedings{zhang-etal-2026-reasoning,
    title = "Reasoning as Gradient: Scaling {MLE} Agents Beyond Tree Search",
    author = "Zhang, Yifei  and
      Yang, Xu  and
      Yang, Xiao  and
      Xian, Bowen  and
      Li, Qizheng  and
      Fang, Shikai  and
      Li, Jingyuan  and
      Wang, Jian  and
      Xu, Minrui  and
      Zhang, Yuge  and
      Liu, Weiqing  and
      Bian, Jiang",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.438/",
    pages = "9013--9038",
    ISBN = "979-8-89176-395-1",
    abstract = "LLM-based agents for machine learning engineering (MLE) predominantly rely on tree search, a form of gradient-free optimization that uses scalar validation scores to rank candidates. As LLM reasoning capabilities improve, exhaustive enumeration becomes increasingly inefficient compared to directed updates, analogous to how accurate gradients enable efficient descent over random search. We introduce Gome, an MLE agent that operationalizes gradient-based optimization. Gome maps structured diagnostic reasoning to gradient computation, success memory to momentum, and multi-trace execution to distributed optimization. Under a closed-world protocol that isolates architectural effects from external knowledge, Gome achieves a state-of-the-art 35.1{\%} any-medal rate on MLE-Bench with a restricted 12-hour budget on a single V100 GPU. Scaling experiments across 10 models reveal a critical crossover: with weaker models, tree search retains advantages by compensating for unreliable reasoning through exhaustive exploration; as reasoning capability strengthens, gradient-based optimization progressively outperforms, with the gap widening at frontier-tier models. Given the rapid advancement of reasoning-oriented LLMs, this positions gradient-based optimization as an increasingly favorable paradigm. We release our codebase and GPT-5 traces at: \url{https://github.com/microsoft/RD-Agent}."
}