English

SubTokenTest: A Practical Benchmark for Real-World Sub-token Understanding

Computation and Language 2026-01-15 v1 Artificial Intelligence

Abstract

Recent advancements in large language models (LLMs) have significantly enhanced their reasoning capabilities. However, they continue to struggle with basic character-level tasks, such as counting letters in words, a problem rooted in their tokenization process. While existing benchmarks have highlighted this weakness through basic character operations, such failures are often dismissed due to lacking practical relevance. Yet, many real-world applications, such as navigating text-based maps or interpreting structured tables, rely heavily on precise sub-token understanding. In this regard, we introduce SubTokenTest, a comprehensive benchmark that assesses sub-token understanding through practical, utility-driven tasks. Our benchmark includes ten tasks across four domains and isolates tokenization-related failures by decoupling performance from complex reasoning. We provide a comprehensive evaluation of nine advanced LLMs. Additionally, we investigate the impact of test-time scaling on sub-token reasoning and explore how character-level information is encoded within the hidden states.

Keywords

Cite

@article{arxiv.2601.09089,
  title  = {SubTokenTest: A Practical Benchmark for Real-World Sub-token Understanding},
  author = {Shuyang Hou and Yi Hu and Muhan Zhang},
  journal= {arXiv preprint arXiv:2601.09089},
  year   = {2026}
}
R2 v1 2026-07-01T09:03:42.427Z