ttok 1.0
Blog · Watch/Read · (gathered 2026-10-14)
- Simon Willison released
ttok 1.0, a tool designed to count and truncate text based on tokens, primarily to update its default tokenizer to align with the latest OpenAI GPT models. - The core change in
ttok 1.0is switching the default tokenizer from GPT-4 to the GPT-5/GPT-6 family. - This update is supported by an experiment by William Liu, which found that GPT-5.x and GPT-6 models (including 5.5, 5.6 Sol/Terra/Luna, 6 Astra/Sol/Luna) likely share the same tokenizer, showing consistent token counts across a test corpus.
- Users of
ttokcan now expect token counts that are consistent with OpenAI’s most recent and upcoming large language models, providing more accurate estimations for LLM interactions.