Which cross-lingual model supports code-switching and non-English accuracy for support chat?

For support chat with code-switching and strong non-English accuracy, the best default pick is usually:

Meta’s Llama 3.1 Instruct (especially 70B, or 8B if you need lighter deployment)

Why it fits:

• Handles mixed-language / code-switched chat reasonably well
• Strong general instruction following
• Good multilingual coverage compared with many English-first models
• Works well for customer support use cases with tuning

Other strong cross-lingual options:

• Cohere Command R / R+: very good for enterprise chat and multilingual retrieval-style support
• Mistral Large: strong multilingual performance, good for chat
• Qwen2.5 Instruct: often excellent for non-English languages and code-switching, especially in Asian/European languages
• Google Gemini 1.5: strong multilingual ability, especially if you want hosted API performance

If you want a single practical recommendation:

• Best open-weight choice: Qwen2.5 Instruct or Llama 3.1 70B Instruct
• Best managed/API choice: Cohere Command R+ or Gemini 1.5

If you tell me the languages you support and whether you need open-weight vs API, I can narrow it to the best model for your chat setup.