Sonic 3.5 is currently in preview on the
sonic-3-latest alias. As with any -latest alias, sonic-3-latest can be updated without notice and is not recommended for production. Pin to a dated sonic-3 snapshot for production traffic.sonic-3-latest. It delivers more natural speech, cleaner audio, dramatically better alphanumeric read-out, and step-change multilingual quality — all while keeping your existing voices and requests working as-is.
Key improvements over sonic-3:
- More natural speech, pacing, and emotional expression, especially on expressive, conversational, and support-style transcripts.
- Cleaner audio quality across all languages and voices.
- Dramatically better alphanumeric read-out - confirmation codes, order numbers, phone numbers, IDs, and emails sound meaningfully more natural across all supported languages.
- Step-change multilingual performance, particularly Hebrew, Japanese, Spanish, Hindi, German, Korean, and French.
- English heteronyms like “read,” “bass,” and “bow” now have more accurate pronunciation in context.
How to try it
- Point your API call or Playground request to the model ID
sonic-3-latest. - Keep your existing voice IDs, request shape, and prompting.
- Send us feedback on any voice or transcript that behaves differently than you expect.
What to know before you switch
- Spell tags work the same way. If you already wrap alphanumerics in
<spell>...</spell>, you don’t need to change anything — you’ll just get better-sounding output. If you use punctuation (e.g. commas, periods, spaced) instead of spell tags, then the recommended format has changed, see prompting tips for details. - Speed and volume controls are temporarily disabled on
sonic-3-latest. If you rely on speed or volume augmentation (including via SSML), stay on datedsonic-3snapshots for now. See Volume, Speed, and Emotion in Sonic 3.5. We believe that Sonic 3.5 has more natural pacing and you may find that you don’t need to use speed control as much when using this model. - Timestamps. If you use end-of-word timestamps for interruption handling, you should not see a meaningful change. If you depend on beginning-of-word timestamps please test carefully and reach out if you see regressions for your use case.
- Pin to a dated snapshot (e.g.
sonic-3) for production traffic.sonic-3-latesttracks the newest release (currently Sonic 3.5) and can change without notice. - Pass a full sentence or clause per request where possible. Very short inputs (single words, fragments) give the model less context to work with and may produce less natural prosody. Please be sure to end transcript sentences with a period. Please see our buffering guide here for more details.
- Keep prompts in the natural written form for your use case. Heavy pre-processing (stripping punctuation, forcing all caps, etc.) generally hurts output quality.