# Cartesia > Real-time speech AI for developers: fast text-to-speech, transcription and voice agents - Category: [Audio & Speech AI](https://finderatlas.com/categories/audio-ai) - Pricing: Freemium - Website: https://www.cartesia.ai - Platforms: Web, API - API: available - Made by: Cartesia, founded 2023, San Francisco, CA - Last updated: 2026-10-19 - FinderAtlas page: https://finderatlas.com/tools/cartesia ## Overview Cartesia is a speech AI platform for developers and enterprises that provides text-to-speech, speech-to-text and managed voice agents through an API. Its Sonic models generate speech with a published time to first audio of about 90 milliseconds, and its Ink models handle streaming transcription with built-in turn detection. Voices can be cloned from a short audio clip, with the docs citing support for 40+ languages and accents. ## Key features - Sonic Text-to-Speech - Ink Speech-to-Text - Managed Agents - Voice Cloning - Pronunciation Dictionaries - Regional API Routing - Telephony Support - Python and JavaScript SDKs - Usage and Access Controls ## Pricing - **Free** ($0): Entry tier for testing text-to-speech and transcription - **Pro** ($5/month): Individuals building with commercial use rights - **Startup** ($49/month): Small teams shipping voice products - **Scale** ($299/month): Higher-volume production workloads - **Enterprise** (Custom): Large organizations with custom requirements ## Best for - Low-latency voice agents, real-time text-to-speech and streaming transcription inside applications ## Who uses it - Developers - Voice AI Engineers - Contact Centre Teams - Enterprises - Product Teams ## Strengths - Published time to first audio of about 90 ms for Sonic 3.6, suited to live conversation. - Free tier with 20K credits per month and a $5 Pro plan with commercial use and instant cloning. - One platform covers text-to-speech, speech-to-text and managed voice agents. - Regional endpoints in the US, EU and APAC, with on-premise and on-device options. - Frequent releases: the changelog shows monthly model, voice and API updates in 2026. - Ranked first on Artificial Analysis speech leaderboards according to the company. ## Limitations - The free tier has no instant voice cloning, only 2 concurrent TTS requests and no commercial license. - Credit-based billing with overage up to $65 per 1M credits can be hard to forecast as usage grows. - Independent reviews note a smaller preset voice library than some narrative-focused competitors. - Developer-oriented; there is no consumer app and no released iOS or Android app. ## Not recommended for - Non-technical users wanting a consumer voiceover app, or teams needing a large library of preset narration voices ## Integrations - Twilio - Anthropic - OpenAI ## FinderAtlas editorial scores (out of 10) - Ease of Use: 7.8/10 - Feature Depth: 8.5/10 - Value for Money: 8.2/10 - Performance & Speed: 9/10 - Documentation & Support: 8.2/10 ## Alternatives - [Tavus](https://finderatlas.com/tools/tavus): Real-time conversational AI video agents with lifelike faces, voice and memory, built via API ([compare](https://finderatlas.com/compare/cartesia-vs-tavus)) - [Otter.ai](https://finderatlas.com/tools/otter-ai): Record, transcribe and summarize meetings, with action items and searchable AI chat ([compare](https://finderatlas.com/compare/cartesia-vs-otter-ai)) - [Wondercraft](https://finderatlas.com/tools/wondercraft): AI studio for turning scripts, documents and ideas into podcasts, audio ads and video ([compare](https://finderatlas.com/compare/cartesia-vs-wondercraft)) - [ElevenLabs](https://finderatlas.com/tools/elevenlabs): Text-to-speech, voice cloning, dubbing and voice agents built on in-house audio models ([compare](https://finderatlas.com/compare/cartesia-vs-elevenlabs)) - [Descript](https://finderatlas.com/tools/descript): Edit video and podcasts by editing their transcript, with AI audio cleanup and clips ([compare](https://finderatlas.com/compare/cartesia-vs-descript)) - [Speechify](https://finderatlas.com/tools/speechify): Listen to documents and web pages with natural AI voices, and dictate by voice ([compare](https://finderatlas.com/compare/cartesia-vs-speechify)) - [Murf AI](https://finderatlas.com/tools/murf): Studio-quality AI voiceovers and text-to-speech API for creators, teams and voice agents ([compare](https://finderatlas.com/compare/cartesia-vs-murf)) - [Fish Audio](https://finderatlas.com/tools/fish-audio): Expressive text-to-speech, voice cloning and speech-to-text with fine-grained emotion control ([compare](https://finderatlas.com/compare/cartesia-vs-fish-audio)) ## Recent updates - 2026-09-01: Ink-2 speech-to-text added Spanish, French, Hindi and Japanese - 2026-08-01: Sonic 3.6 became generally available with Odia and Urdu support, and Managed Agents launched with hosted LLMs and bui... - 2026-05-01: Ink-2 launched, initially English-only, alongside Sonic 3.5 general availability - 2026-01-01: US, EU and APAC routing launched ## FAQ **What is Cartesia Sonic?** Sonic is Cartesia's text-to-speech model family. The current generation, Sonic 3.6, became generally available in August 2026 and is listed with about 90 ms time to first audio. **Does Cartesia have a free plan?** Yes. The Free plan costs $0 and includes 20K credits per month, text-to-speech and speech-to-text, but no instant voice cloning. A commercial use license starts on the $5 Pro plan. **Can I clone a voice with Cartesia?** Instant voice cloning is available on Pro and above, and professional voice clones on Startup, Scale and Enterprise. The docs state cloning works from a 10-second clip across 40+ languages and accents. **How much does a voice agent cost?** Agent call duration is $0.06 per minute, plus $0.014 per minute for telephony on a Cartesia-provided number. LLM usage was free for UI-created agents until October 1, 2026. **Can Cartesia run on-premise or on-device?** The website lists cloud deployment through regional API endpoints, as well as on-premise and on-device options. Details are not published on the pages reviewed. **Which older models are being retired?** The changelog says Sonic-2, Sonic-turbo and Sonic-3-2025-10-27 will be retired after October 20, 2026, and Line SDK agents stop being hosted on December 1, 2026.