Developer Simon Willison released ttok 1.0 on Oct. 9, switching his token-counting tool's default tokenizer from GPT-4 to the one used by GPT-5, betting that GPT-6 uses the same one, according to his blog.
Willison made the change after noticing that ttok 0.4, released a day earlier, still defaulted to the outdated GPT-4 tokenizer, he wrote. OpenAI has not confirmed that GPT-6 shares a tokenizer with the GPT-5 family, but Willison cited testing by developer William Liu showing that all seven current GPT models report an identical count of 44,794 tokens on the same test data.
ttok works with uvx, Python's tool runner, letting developers count tokens in a file without installing anything first, Willison wrote.
Token counts drive both API costs and context-window limits, so a popular counting tool quietly defaulting to the wrong tokenizer would throw off bills and budgets without producing any error message.