Multilingual support

All Mira models are trained on multilingual data and support over 50 languages. Special attention is given to Russian and English, which are supported at native-speaker level.

Supported languages

Priority languages (highest quality)

LanguageCodeQualityNotes
RussianruExcellentNative level, all models
EnglishenExcellentNative level, all models

Well-supported languages

The following languages are supported at a high level with minimal quality loss compared to priority languages:

  • EuropeanGerman, French, Spanish, Italian, Portuguese, Dutch, Polish, Ukrainian, Czech, Swedish
  • AsianChinese (Simplified/Traditional), Japanese, Korean, Vietnamese, Thai, Indonesian
  • Middle EasternArabic, Turkish, Hebrew, Persian
  • IndianHindi, Bengali, Tamil, Telugu

Performance across languages

Mira models are optimized to maintain quality when working with non-English languages. Here is approximate performance by language:

Language group% of English qualityTokenization
Russian98%Optimized
European (Latin)95%Efficient
Chinese / Japanese / Korean92%Efficient
Arabic / Hebrew88%Good
Indian languages85%Good
Russian has optimized tokenization, meaning fewer tokens are consumed for the same text length compared to other Cyrillic languages.

Best practices for non-English prompts

  • Write prompts in the target languagefor best Russian output quality, ask in Russian rather than asking in English to respond in Russian
  • Be specificspecify desired style and formality in the target language
  • System messagesset the system message in the target language to maintain context
  • Avoid mixingfor maximum quality, avoid mixing languages in a single prompt (except for technical terms)

Russian language examples

Russian prompt
curl https://api.vmira.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-mira-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mira",
    "messages": [
      {
        "role": "system",
        "content": "Ты — опытный технический писатель. Отвечай на русском языке, используя профессиональный, но доступный стиль."
      },
      {
        "role": "user",
        "content": "Объясни, что такое REST API, простыми словами."
      }
    ]
  }'

Translation

Mira excels at translation between any supported languages. For best results, provide context and specify the target audience.

Translation example
{
  "model": "mira",
  "messages": [
    {
      "role": "system",
      "content": "You are a professional translator. Translate from Russian to English, preserving the style and tone of the original."
    },
    {
      "role": "user",
      "content": "Translate: \"Artificial intelligence opens new horizons in science and technology.\""
    }
  ]
}

Language detection

Mira automatically detects the language of input text and can switch languages within a single conversation. This is useful for applications serving users from different countries.

Mixed-language conversations

Mira can handle conversations where the user switches between languages. The model responds in the same language the question was asked in, unless instructed otherwise.

Language switching in conversation
{
  "model": "mira",
  "messages": [
    { "role": "user", "content": "What is machine learning?" },
    { "role": "assistant", "content": "Machine learning is a subset of AI..." },
    { "role": "user", "content": "А теперь объясни это на русском." },
    { "role": "assistant", "content": "Машинное обучение — это подраздел ИИ..." }
  ]
}
Tip: for multilingual apps, use a system message with the instruction "Respond in the same language as the question." This ensures automatic switching.