> First, Opus 4.7 uses an updated tokenizer that improves how the model processes text

wow can I see it and run it locally please? Making API calls to check token counts is retarded.