I Measured Claude 4.7's New Tokenizer. Here's What It Costs You.
Summary
Anthropic's Claude Opus 4.7 tokenizer uses more tokens per prompt, leading to 20–30% higher per-session costs despite the same sticker price. Real tests show a weighted average token increase around 1.325x for English and code, with higher ratios for certain file types; the article also notes a small improvement in strict instruction following per IFEval and explains caching implications. The content helps readers understand cost implications, when to expect bigger bills, and how content type influences value.