“Leaker Leo claims that GPT-6 will be based on a new, significantly larger pre-training dataset compared to the 4 trillion token base used for GPT-5.5 and GPT-5.6.”