CharBench - Character-level benchmark and analysis suite for LLMs.

Code is available in the GitHub Repo.

CharBench is a large-scale benchmark for studying tokenization and character-level behavior in modern language models.

For complete details on data curation and evaluation, see the paper.

Usage

from datasets import load_dataset
ds = load_dataset("omriuz/CharBench")

Citation

If you use CharBench, please cite:

@misc{uzan2025charbenchevaluatingroletokenization,
      title={CharBench: Evaluating the Role of Tokenization in Character-Level Tasks}, 
      author={Omri Uzan and Yuval Pinter},
      year={2025},
      eprint={2508.02591},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2508.02591}, 
}