Overview

Claude demonstrates strong multilingual capabilities across a wide range of languages, with particularly robust performance in zero-shot tasks. The model shows consistent performance across both widely-spoken languages like English, Chinese, and Spanish, as well as languages with fewer digital resources like Swahili and Yoruba.

Note that Claude is highly performant in languages not benchmarked below. We encourage customers to evaluate any languages as needed for their specific use cases.

Performance data

Below are the zero-shot chain-of-thought evaluation scores for Claude 3.5 models across different languages:

These metrics are based on MMLU (Massive Multitask Language Understanding) English test sets that were translated into 14 additional languages by professional human translators, as documented in OpenAI’s simple-evals repository. The use of human translators for this evaluation increases confidence in the accuracy of the translations, especially for low-resource languages like Yoruba.

LanguageClaude 3.5 Sonnet (New)Claude 3.5 Haiku
English90.1%81.5%
French86.7%77.7%
Spanish87.3%77.1%
Portuguese (Brazil)86.5%77.1%
Italian86.1%77.4%
German84.7%75.4%
Chinese (Simplified)83.6%74.1%
Indonesian84.7%74.3%
Japanese83.5%74.0%
Korean83.6%72.6%
Arabic83.3%69.0%
Hindi80.5%65.3%
Bengali77.4%59.4%
Swahili75.6%52.7%
Yoruba58.5%37.6%

Best practices

When working with multilingual content:

  1. Provide clear language context: While Claude can often detect the target language, explicitly stating the desired input/output language improves reliability. If you find that Claude is struggling with fluency, you can also prompt Claude to use “idiomatic speech as if it were a native speaker.”
  2. Use native scripts: Submit text in its native script rather than transliteration for best results
  3. Consider cultural context: Remember that effective communication often requires cultural and regional awareness beyond pure translation

Language support considerations

  • Claude can process input and generate output in most world languages that use standard Unicode characters
  • Performance may vary between languages, with generally stronger capabilities in widely-spoken languages