Study: Byte-Level LLMs Can Overcome Subword Tokenization Limits

Researchers from LMU, the Allen Institute for AI, Cambridge, the University of Washington and Imperial College London published a Nature method that converts existing LLMs into byte-level models without retraining from scratch, using under 1% of the usual compute. The byteified models largely keep original performance, add byte-level benefits, outperform prior same-size byte-level LLMs and handle character-level tasks like spelling backwards far better.

• Bender says AI's 'existential' risk is 'fake'2026-10-06
• Economist Diane Coyle: AI Will Ultimately Benefit the General Public2026-09-19
• UW professor: AI extinction unlikely2026-09-17