Compare UBE with released MYTE and Bit-level BPE at the model level

Conduct direct model-level comparisons between Universal Byte-Level Encoding and the released 256-symbol MYTE representation and Bit-level BPE under matched training, exposure, architecture, and evaluation conditions.

Background

The paper evaluates a derived 32K MYTE+BPE variant and includes tokenizer-level or representation-level comparisons with several alternatives, but it does not directly evaluate the released 256-symbol MYTE model configuration or Bit-level BPE at the model level.

The authors explicitly defer these direct comparisons. Such experiments would clarify whether UBE’s observed token-budget and language-model behavior is competitive with these alternative byte representations when evaluated under comparable model-training conditions.

References

Direct model-level comparisons with the released 256-symbol MYTE itself [Limisiewicz et al., 2024] and Bit-level BPE [Moon et al., 2025] remain future work.

— Universal Byte-Level Encoding: UTF-8/UTF-16 Routing to Reduce Cross-Script Token-Budget Disparities  (2610.01984 - Kim et al., 1 Oct 2026) in Conclusion, “Limitations”; Appendix H.4, “Comparison scope”