Anonymous request-token comparisons from Opus 4.6 and Opus 4.7

https://tokens.billchambers.me/leaderboard

Article

  • Leaderboard comparing token counts for identical prompts across Opus 4.6 and 4.7.
  • Measures tokenizer change in isolation via the token-counting API.
  • Shows 4.7 consumes significantly more input tokens for the same prompts.

Discussion

  • andai: 4.7 produces fewer output tokens, so total cost may actually be lower—need full cost comparison.
  • Users hitting 5-hour limits in 2 hours with 4.7; 300-line HTML site exhausted weekly limit.
  • hereme888 cited Artificial Analysis benchmark: 4.7 costs ~11% less than 4.6 despite higher input token counts.
  • Divided community: some find 4.7 clearly better, others switched to Codex or stuck with 4.5.

Discuss on HN


Type Link
Added Apr 20, 2026
Modified Apr 20, 2026