Can LLMs Beat Classical Hyperparameter Optimization Algorithms?

Source: Hacker News

View PDF HTML (experimental) Abstract:The autoresearch repository enables an LLM agent to optimize hyperparameters by editing training code directly. We use it as a testbed to compare classical HPO algorithms against LLM-based methods on tuning the hyperparameters of a small language model under a fixed compute budget. When defining a fixed search space over autoresearch, classical methods such as CMA-ES and TPE consistently outperform LLM-based agents, where avoiding out-of-memory failures matters more than search diversity. Allowing the LLM to directly edit source code narrows the gap to the classical methods but does not close it, even with frontier models available at the time of writing such as Claude Opus 4.6 and Gemini 3.1 Pro Preview. We observe that LLMs struggle to track optimization state across trials. In contrast, classical methods lack the domain knowledge of LLMs. To combine the strengths of both, we introduce Centaur, a hybrid that shares CMA-ES’s interpretable internal state, including mean vector, step-size, and covariance matrix, with an LLM. Centaur achieves the best result in our experiments, and a 0.8B LLM already suffices to outperform all classical and pure LLM methods. Unconstrained code editing requires larger models to be competitive with classical methods. We further analyze search diversity, model scaling from 0.8B to frontier models, and ablate the fraction of LLM-proposed trials in Centaur. All in all, our results suggest that LLMs are most effective as a complement to classical optimizers, not as a replacement.

Code is available at this https URL & interactive demo at this https URL.

      Subjects:
      
        Machine Learning (cs.LG); Machine Learning (stat.ML)
    
      Cite as:
      [arXiv:2603.24647](https://arxiv.org/abs/2603.24647) [cs.LG]
    
    
       
      (or 
          [arXiv:2603.24647v5](https://arxiv.org/abs/2603.24647v5) [cs.LG] for this version)
      
    
    
       
                    [https://doi.org/10.48550/arXiv.2603.24647](https://doi.org/10.48550/arXiv.2603.24647)
          
          
              arXiv-issued DOI via DataCite

        
      
    


  

Submission history

From: Fabio Ferreira [view email]
[v1] Wed, 25 Mar 2026 17:29:40 UTC (1,874 KB)

        **[[v2]](https://arxiv.org/abs/2603.24647v2)**
    Sun, 29 Mar 2026 18:46:53 UTC (2,456 KB)

        **[[v3]](https://arxiv.org/abs/2603.24647v3)**
    Sat, 4 Apr 2026 10:33:34 UTC (3,843 KB)

        **[[v4]](https://arxiv.org/abs/2603.24647v4)**
    Mon, 13 Apr 2026 21:59:37 UTC (3,768 KB)

**[v5]**
    Fri, 17 Apr 2026 18:50:51 UTC (3,905 KB)