[BUG] Model omits diacritical marks (accents) in non-English languages unless explicitly requested
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
Any prompt written in a language that uses diacritical marks (e.g., Portuguese, French, Spanish). For example, asking Claude Code to write comments, commit messages, or respond in Portuguese.
When responding in languages that use diacritical marks (accents, tildes, cedillas, etc.), the model consistently omits them from words that require them. For example:
"nao" instead of "não"
"codigo" instead of "código"
"voce" instead of "você"
"tambem" instead of "também"
"e" instead of "é"
Accents are only placed correctly when the user explicitly asks the model to use them. This happens across all types of output: responses, code comments, commit messages, and file content.
What Should Happen?
The model should use correct diacritical marks by default when writing in any language that requires them, without the user needing to explicitly request proper orthography. Omitting accents changes meaning in many languages (e.g., in Portuguese, "e" means "and" while "é" means "is").
Error Messages/Logs
Steps to Reproduce
- Open a project that uses internationalization (i18n) with translation files (e.g., JSON locale files).
- Ask Claude Code to add or modify UI elements that require generating or updating translation strings in a non-English language that uses diacritical marks (e.g., Portuguese).
- Example prompt: "Add a settings page with all labels in Portuguese"
- Observe the generated translation/locale files — diacritical marks (accents, tildes, cedillas) will be missing from words that require them.
- Example: "notifications": "Notificacoes" instead of "notifications": "Notificações"
- Example: "schedule": "Agendamento de publicacao" instead of "Agendamento de publicação"
- The only way to get correct accents is to explicitly instruct: "Use correct Portuguese accents in all strings."
Claude Model
Opus
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.90 (Vs Code extension)
Platform
Anthropic API
Operating System
Other Linux
Terminal/Shell
Other
Additional Information
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗