> but the gap is still massive even if you switch to comparing the number of tokens.

Token doesn't have the same meaning at OpenAI vs Anthropics, they use different tokenizer. How can this be used for comparison?