claude-code - 💡(How to fix) Fix Opus 4.7 tokenizer change creates systematic pricing disadvantage for Latin-script non-English users

The Claude Opus 4.7 tokenizer change introduces a structural pricing disadvantage for Latin-script non-English users (Hungarian, Polish, Czech, Romanian, etc.).

The new tokenizer was optimized to reduce token counts for CJK and Arabic scripts — a legitimate improvement. However, the tradeoff is that the same content in Hungarian now produces ~1.3–1.47× more tokens compared to Opus 4.6, while the per-token price remains unchanged.

This means Hungarian (and similarly affected) users are effectively paying 30–47% more per task than English or CJK users for the same information content. This is not a minor inconvenience — it's a systematic pricing disparity baked into the model layer, invisible to users who don't audit their token counts.

Requested action:

1. Publish per-language tokenization efficiency benchmarks for Opus 4.7 vs 4.6 so affected users can make informed decisions.
2. Consider language-aware token allocation adjustments for subscription plans (as already requested for CJK users in #26401).
3. If the tokenizer tradeoff is permanent, provide a clear migration path or pricing adjustment for disproportionately affected language groups.

Affected users have no visibility into this cost increase — it's hidden behind an "unchanged price" narrative while the effective cost per task has meaningfully increased.

Related: #26401