Supporting 100+ languages is a meaningful achievement. But it's worth being honest: translation quality isn't uniform across all language pairs. Here's what you should know.

Tier 1: Excellent (Most Use Cases)

These language pairs have the most training data and perform reliably well:

  • English ↔ Spanish
  • English ↔ French
  • English ↔ German
  • English ↔ Japanese
  • English ↔ Chinese (Simplified)
  • English ↔ Korean
  • English ↔ Portuguese
  • English ↔ Italian

For these pairs, real-time translation handles natural conversation, idioms, and moderate technical vocabulary with high accuracy.

Tier 2: Very Good (Everyday Use)

  • English ↔ Russian
  • English ↔ Arabic
  • English ↔ Hindi
  • English ↔ Dutch
  • English ↔ Polish
  • English ↔ Turkish
  • Chinese ↔ Japanese
  • Spanish ↔ Portuguese

These pairs work well for most calls. Occasional awkward phrasing may appear, but meaning is consistently conveyed.

Tier 3: Good (With Some Care)

Less common language pairs or those with significant structural differences may require slightly slower speech and simpler sentences for best results. These still work well for practical calls — just avoid dense technical language.

Tips for Any Language Pair

Speak in complete sentences. Fragments are harder to translate accurately.

Pause between thoughts. Real-time systems process in chunks — brief natural pauses help.

Avoid idioms when precision matters. "Let's touch base" or "ballpark figure" can confuse translation models, especially into languages where these phrases don't have direct equivalents.

Use the transcript. If something sounds off in the audio, the text transcript often shows more accurate translation.

Translation quality improves continuously. Models are retrained regularly, and the language pairs in Tier 2 and 3 today will be Tier 1 quality within the next year.