Skip to main content
We’re moving to updating the Changelog monthly, with changes across the model, voices, UI and API changes.

​
Model changes

  • Sonic-3 model versioning scheme introduced
    • New preview track: sonic-3-latest (continuous updates for early access and feedback).
    • Stable track: sonic-3 always points to the most recent stable release.
    • Immutable dated snapshots: sonic-3-YYYY-MM-DD never change.
    • Details: Continuous updates and model snapshots
  • Promotion to stable checkpoint: sonic-3-2026-01-12
    • sonic-3-latest changes graduated to sonic-3 and the stable snapshot sonic-3-2026-01-12.
    • Included improvements:
      • Consistent speed & volume for more predictable output without losing expressiveness.
      • Custom IPA pronunciations with stronger IPA adherence.
      • Hindi prosody improvements (more natural rhythm, intonation, pause handling).
      • Korean prosody/intonation improvements.

​
Voice changes

  • Featured Voices launched (curated voice set)
    • Introduced Featured Voices as a curated selection of best-performing voices.
    • Launched with 30+ featured voices, with ongoing additions.
    • Example voices:
  • Voice Library additions
    • December updates: 25 new voices across 6 languages
      • 12 English
      • 6 Hindi
      • 4 Arabic
      • 1 Spanish
      • 1 Japanese
      • 14 selected as featured voices
    • January updates: 9 Spanish voices across accents
      • Mexican, Colombian, Castilian
      • 5 selected as featured voices

​
UI changes

  • Voice library usability improvements (Playground)
    • Easier to test voices using your own scripts.
    • Ability to call an agent to hear what end users will hear.
    • Ability to call each voice
  • One-click feedback on TTS Playground
    • Added Report Issue button to send feedback with instant context.
  • Mini voice picker on TTS page
    • Inline dropdown showing recently used and saved voices.
    • Quick switching without opening the full voice picker dialog.
  • PVC UI + reliability improvements
    • UX improvements: better loading skeletons and error messages
    • Reliability fixes for PVCs: PVCs now work better with very large datasets and datasets with lots of silence

​
Line

  • Line SDK v0.2 (developer experience + features)
    • Repo: https://github.com/cartesia-ai/line
    • Highlights:
      • Improved developer experience
      • Built-in handling of long-running tool calls
      • Committed turns: agents can be made aware of what they actually said if interrupted
      • Improved call performance: better turn-taking and transcription quality
  • Regionalization (routing)
    • Calls routed to US, EU, APAC based on origin for improved latency and reliability.
  • Parameterized outbound calls
    • Pass data to enable customized agent behavior per phone call.
    • Docs
  • Pronunciation dictionaries (Line)
    • Pronunciation dictionaries can now be passed to agents: specify custom pronunciations for hard-to-say words like proper nouns and domain terms.
    • Docs