4.7 Token Surge

Claude Opus 4.7's official documentation states it may use 1x~1.35x more tokens compared to the previous model, but separate measurements showed even higher figures.

• The same content was fed into 4.6 and 4.7 respectively using Anthropic's free token counter POST /v1/messages/count_tokens for comparison.
• The results showed token usage increasing by up to 1.47x.
• Examples:
  • Technical documentation (English, 2,541 characters): 478 → 704 tokens (1.47x)
  • Actual CLAUDE.md file (about 5KB, 5,000 characters): 1,399 → 2,021 tokens (1.445x)
• Characters per token also dropped, from 4.33 → 3.60 for English and 3.66 → 2.69 for TypeScript, meaning the same text is being split into smaller pieces.

In other words, Opus 4.7's new tokenizer, alongside performance improvements, may increase actual billing and context consumption, meaning workflows with long-text inputs need to recalculate costs and context headroom.