Real-time voice agents
Use streaming recognition, built-in turn detection, and low-latency speech generation for natural back-and-forth conversations.
Independent tool overview
Sonic 3.6 & Ink 2 at a glance
Sonic 3.6 and Ink 2 are Cartesia's real-time text-to-speech and speech-to-text models for building low-latency voice agents through one API platform.
Visit the official Sonic 3.6 & Ink 2 site ↗
- Product type
- Real-time TTS and STT APIs
- Developer
- Cartesia
- Current TTS model
- Sonic 3.6
- Current STT model
- Ink 2
- Sonic languages
- 44
- Ink 2 languages
- English
- Starting price
- Free
- Paid plans
- From $5 per month
- Last reviewed
- August 29, 2026
Overview
What Sonic 3.6 & Ink 2 is
Cartesia pairs Sonic 3.6 for text-to-speech with Ink 2 for streaming speech recognition. Together they cover the audio input and output layers of a real-time voice agent without requiring separate speech vendors.
This listing originally covered Sonic 3.5 and Ink 2. Sonic 3.6 is now the generally available TTS model and is fully backward-compatible with 3.5, so new projects should evaluate the current model ID, sonic-3.6.
Sonic 3.6 supports 44 languages, voice cloning, streaming, locale-aware text normalization, and stable dated snapshots. Ink 2 is currently English-only and focuses on low-latency transcription, structured data, and built-in turn detection.
Cartesia advertises sub-90ms TTS and 100ms transcript latency for the combined real-time stack. Those are vendor claims; production latency will also depend on geography, network conditions, buffering, audio transport, the language model, and application logic.
Pricing uses shared monthly credits. TTS is roughly one credit per input character, while Ink 2 costs three credits per second of audio, including silence. Teams should test realistic calls and set overage controls before forecasting costs from the headline plan price.
Use cases
Who Sonic 3.6 & Ink 2 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Customer-support automation
Handle confirmation codes, dates, phone numbers, emails, names, and conversational pacing in support-style calls.
Multilingual speech output
Generate speech across 44 languages with localized pronunciation and locale-aware rendering.
Developers consolidating vendors
Connect both speech recognition and synthesis through one provider, account, credit pool, and integration surface.
Voice cloning projects
Create instant or professional voice clones for approved speakers and supported use cases.
Latency-sensitive applications
Build phone, web, mobile, or avatar experiences where transcript and audio delays directly affect conversation quality.
Capabilities
Core Sonic 3.6 & Ink 2 features
Sonic 3.6 streaming TTS
Streams expressive synthesized speech through bytes, SSE, or WebSocket endpoints.
44-language speech generation
Covers major European and Asian languages plus Odia and Urdu added in Sonic 3.6.
Context-aware delivery
Adjusts pacing and intonation from the transcript and handles disfluencies such as 'uhm' and 'hmm' naturally.
Structured-text pronunciation
Reads confirmation codes, phone numbers, dates, emails, and heteronyms with less manual preprocessing.
Locale support
Uses locale codes such as en-GB to interpret dates and other localized formats correctly.
Voice cloning
Supports instant cloning and plan-dependent professional voice cloning, with older professional clones working on Sonic 3.6.
Ink 2 streaming STT
Transcribes English audio in real time for voice-agent and push-to-talk workflows.
Native turn detection
Emits start, update, eager-end, resume, and end events without requiring a separate voice-activity detector.
Stable model snapshots
Lets production teams pin dated Sonic releases while testing upcoming behavior through a preview alias.
Official SDKs and integrations
Provides Python and JavaScript libraries plus documented integrations with voice-agent and telephony frameworks.
Enterprise data controls
Offers zero data retention for eligible inference requests and dedicated regional deployments on Enterprise.
Process
How the Sonic 3.6 & Ink 2 workflow works
Step 1
Choose the speech layers
Decide whether the application needs Sonic TTS, Ink STT, or both; confirm that English-only recognition fits the input audience.
Step 2
Start with a representative test set
Collect real transcripts and audio covering accents, background noise, names, numbers, interruptions, silence, and difficult pronunciations.
Step 3
Create a server-side API key
Keep permanent credentials off clients and use Cartesia's client authentication approach where browser or mobile access is required.
Step 4
Integrate realtime audio
Stream microphone audio to Ink 2 and route agent text back through Sonic 3.6 using the appropriate WebSocket or streaming endpoint.
Step 5
Tune turn handling
Use Ink 2's turn lifecycle events to decide when the agent should listen, prepare a response, interrupt, or resume.
Step 6
Select and test voices
Compare several voices with real scripts and verify expressiveness, pronunciation, pace, and consistency across target languages.
Step 7
Pin production versions
Use a dated Sonic snapshot when repeatable behavior matters, and evaluate stable alias changes before rolling them into critical workflows.
Step 8
Measure end-to-end performance
Track time to transcript, turn-end accuracy, model response time, time to first audio, interruptions, word error rate, and call completion.
Step 9
Control spend and data
Monitor credits, decide whether to enable overages, set concurrency expectations, and evaluate Enterprise retention or regional requirements.
Cost
Sonic 3.6 & Ink 2 pricing and free plan
Every plan includes a shared monthly credit balance. Approximate TTS minutes and STT hours assume the entire allowance is used on that one capability, so a production stack using both will split the credits. Sonic TTS costs roughly one credit per character; Ink 2 realtime STT costs three credits per second, including silence.
Free
$0 per month
For prototypes and light evaluation with 20,000 monthly credits.
- About 27 Sonic 3.6 minutes if all credits go to TTS
- About 1 hour 51 minutes of Ink 2 if all credits go to STT
- 2 concurrent TTS requests
- 8 concurrent STT requests
Pro
$5 per month
Entry paid plan with commercial-use rights and 100,000 monthly credits.
- About 133 TTS minutes or 9 hours 16 minutes of Ink 2
- 3 concurrent TTS requests
- 12 concurrent STT requests
- Instant voice cloning
Startup
$49 per month
For growing applications with 1.25 million monthly credits and organization features.
- About 1,667 TTS minutes or 115 hours 44 minutes of Ink 2
- 5 concurrent TTS requests
- 20 concurrent STT requests
- Professional voice cloning
Scale
$299 per month
For higher-volume production with 8 million monthly credits.
- About 10,667 TTS minutes or 740 hours 44 minutes of Ink 2
- 15 concurrent TTS requests
- 60 concurrent STT requests
- Priority support and higher concurrency
Enterprise
Custom
For negotiated volume, security, compliance, and deployment requirements.
- Custom credits and concurrency
- DPAs and BAAs
- SSO and security reviews
- Zero data retention and regional deployments available
Pricing checked . Check current pricing at the source ↗
Assessment
Sonic 3.6 & Ink 2 strengths and limitations
Where it stands out
- Covers both speech input and output through one platform
- Designed specifically for low-latency, interruptible voice conversations
- Sonic 3.6 supports 44 languages and locale-aware output
- Ink 2 includes native turn detection rather than requiring a separate VAD
- Handles common structured strings such as phone numbers, dates, emails, and confirmation codes
- Dated Sonic snapshots give production teams a stable versioning option
- Free plan makes realistic prototyping possible before committing
- Official SDKs, realtime endpoints, and integrations support common agent stacks
- Enterprise customers can request zero data retention and regional inference
What to consider
- Ink 2 currently supports English only even though Sonic 3.6 can speak 44 languages
- Sonic 3.5 is no longer the newest TTS model, so old comparisons and model IDs can become stale
- Headline latency claims do not include the full network, LLM, application, or telephony path
- Credit pricing makes per-call cost less obvious than a single per-minute rate
- Ink 2 bills every second of streamed audio, including silence
- Approximate included minutes and hours cannot both be consumed from the same shared credits
- Realtime Ink 2 is not yet available through the batch transcription endpoint
- Voice quality and transcription accuracy still vary by speaker, accent, audio equipment, noise, and script
- Production teams need their own evaluation set rather than relying only on public leaderboards
- Voice cloning requires verified consent and controls against impersonation or deceptive use
- Zero data retention and dedicated regional deployments require Enterprise
- Overages can create unplanned spend if usage monitoring and limits are not configured
Compare
Sonic 3.6 & Ink 2 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Marketing
ElevenLabs
Consider ElevenLabs for a broad voice platform spanning expressive TTS, cloning, dubbing, and conversational agents.
Explore ElevenLabs →Content Creator
AssemblyAI
Choose AssemblyAI when speech recognition and audio intelligence matter more than sourcing TTS and STT from one vendor.
Explore AssemblyAI →Miscellaneous
TADA by Hume AI
Consider TADA when an open-source TTS model and tight text-audio alignment are more important than a managed full-stack voice API.
Explore TADA by Hume AI →Questions
Sonic 3.6 & Ink 2 FAQs
What are Sonic 3.6 and Ink 2?
Sonic 3.6 is Cartesia's current text-to-speech model, while Ink 2 is its streaming speech-to-text model with built-in turn detection. They can form the speech layer of a real-time voice agent.
What happened to Sonic 3.5?
Sonic 3.6 is now generally available and Cartesia says it is fully backward-compatible with Sonic 3.5. New projects should evaluate sonic-3.6, while existing production teams can migrate with their normal regression tests.
How many languages does Sonic 3.6 support?
Sonic 3.6 supports 44 languages. It added Odia and Urdu and also includes locale codes for localized handling of formats such as dates.
How many languages does Ink 2 support?
Ink 2 is currently listed as English-only. Teams needing multilingual transcription should evaluate another Cartesia STT model or a different provider.
How much does Cartesia cost?
Plans start at $0, followed by Pro at $5 per month, Startup at $49, Scale at $299, and custom Enterprise pricing. Each plan includes credits shared across model usage.
How are Sonic and Ink usage billed?
Standard Sonic TTS is approximately one credit per input character. Realtime Ink 2 costs three credits per second of audio, including silence. Only successful requests consume credits.
Does Ink 2 require separate voice activity detection?
No. Its turn-detection endpoint emits a lifecycle of turn events so the application can identify when a speaker starts, pauses, resumes, and ends.
Should production apps use a stable alias or dated Sonic snapshot?
Use the stable sonic-3.6 alias if automatic stable updates are acceptable. Pin a dated snapshot when you need repeatable behavior and want to evaluate every model change before rollout.
Does Cartesia offer zero data retention?
Enterprise customers can enable zero data retention for eligible TTS and STT inference payloads. It does not cover voice cloning workflows, and operational metadata is still retained.
Bottom line
Our Sonic 3.6 & Ink 2 verdict
Cartesia is a strong fit for teams that want a tightly integrated, low-latency voice stack and can work within Ink 2's English-only recognition. Sonic 3.6 offers broad multilingual output and practical production versioning, while Ink 2's turn events simplify conversational orchestration. The deciding test should be an end-to-end pilot using real callers, realistic silence, noisy audio, and a full cost model—not the model latency claim alone.
Visit Sonic 3.6 & Ink 2 website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
DiffusionGemma
DiffusionGemma - Google's open diffusion model that can quadruple text generation speed
Antares
Antares - Cisco's small, open security models that scan code locally
Gemini 3.5 Live Translate
Gemini 3.5 Live Translate - Google's real-time voice model for live translation across 70+ languages
.jpeg)
Lyria 3.5
Lyria 3.5 - Google's new music model with more realistic vocals
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.