Cartesia Ink 2
Cartesia
Ink 2 is Cartesia's streaming speech-to-text model for real-time voice agents, exposed through the API as model id ink-2. It transcribes English only and has turn detection built in: rather than a separate VAD it emits a lifecycle of turn events (turn.start, turn.update, turn.eager_end, turn.resume, turn.end) driven by semantic endpointing that decides a turn is over by meaning rather than silence. Cartesia reports a time-to-final-transcript of 0.1s and accuracy on structured entities such as phone numbers, dates and email addresses.
Strengths
- Semantic endpointing decides turn boundaries by meaning rather than silence, with no external VAD required
- Emits a full turn lifecycle (turn.start, turn.update, turn.eager_end, turn.resume, turn.end), so an LLM can start generating on turn.eager_end
- Time-to-final-transcript of 0.1s reported by Cartesia
- 8% WER on AppTek, a 14-accent English call-centre benchmark, against 10% for Deepgram Flux and 12% for ElevenLabs Scribe v2
- 6.5% WER on Cartesia's internal benchmark of live voice-agent audio, against 9.2% for ElevenLabs Scribe v2 and 9.4% for Deepgram Flux
- Transcribes structured data such as phone numbers, dates and email addresses without special prompting
Caveats
- English only: the model table lists `en` as the sole supported language, and Cartesia says only that 'multilingual support is on the way'.
- Both published WER figures come from benchmarks Cartesia selected - AppTek and its own internal production set - with no independent evaluation cited.
- Closed model, API access only; no weights are published and no self-hosting path is offered.
- No per-minute STT price is published: hours are bundled into subscription tiers, and concurrency is capped by plan at 8 requests on Free rising to 60 on Scale.
- The predecessor ink-whisper is now filed under 'Older STT Models'; the Ink 2 page gives no accuracy or migration comparison between the two.
- Vendor
- Cartesia
- License
- Proprietary
- Release Date
- 2026-05-22
- Modalities
- audiotext
| Benchmark | Score | Max |
|---|---|---|
| AppTek multi-accent English call-centre WER | 8.0 | 100.0 |
| Cartesia internal production voice-agent audio WER | 6.5 | 100.0 |
Capabilities
Resources
Pricing
No standalone per-minute STT rate is published; Ink-2 hours are bundled into Cartesia's subscription plans (Free $0/mo, Pro $5/mo, Startup $49/mo, Scale $299/mo, Enterprise custom) with per-plan concurrent-request caps of 8, 12, 20 and 60.