Cartesia vs ElevenLabs: Voices, Speech-to-Text and Agents

ElevenLabs has the top two voices on the Speech Arena and a more developed agent platform. Cartesia's Sonic 3.6 ranks fourth, its Ink 2 speech-to-text detects turns, and Managed Agents lists $0.03 a minute. To use both, Dograh, our open-source AI voice agent platform, runs each on your own keys, self-hosted or on Dograh Cloud.

Key Takeaways

• Eleven v4 Turbo and Eleven v4 rank first and second, Sonic 3.6 fourth.
• Both vendors offer residency, a BAA and private deployment on Enterprise.
• Inside Dograh, one agent can mix a vendor's voice with the other's transcription.

Choosing between Cartesia and ElevenLabs is really three choices, because each vendor sells voices, speech to text and an agent platform. We set them side by side layer by layer, say who each one suits and show how a single agent in Dograh can use both. It is part of our guide to Cartesia alternatives.

Cartesia vs ElevenLabs at a glance

Both vendors build their own speech models and now sell a hosted agent platform on top. The table compares the three layers, with list prices from each vendor's own pages.

Voices: Sonic 3.6 vs Eleven v4

On the Artificial Analysis Speech Arena, where listeners vote blind, Eleven v4 Turbo ranks first, Eleven v4 second and Sonic 3.6 fourth. Further down, v3 Conversational and Sonic 3.5 sit around 20th, and Flash v2.5 falls outside the top 50. The top places sit close together on that board, while the lower entries trail well behind.

Prices are set differently. Cartesia charges about 1 credit a character from a monthly plan, which our arithmetic puts at roughly $0.037 to $0.050 per 1,000 characters, depending on the plan. ElevenLabs' API pricing lists $0.04 per 1,000 characters for Flash, Turbo and v4 Turbo, and $0.08 for v4.

Both clone voices in two tiers. Cartesia's Instant Clone needs about 10 seconds of audio, and its Professional clone needs at least 30 minutes and sounds native only in the language it was trained on. Sonic 3.6 adds a locale setting for regional accents and switches between Hindi and English in one sentence.

Speech-to-text: Ink vs Scribe

Cartesia's Ink 2 covers English, French, Hindi, Japanese and Spanish, and it has turn detection built in, sending events as the caller starts, pauses and finishes. Cartesia puts it at $0.39 an hour on its Scale plan. The older ink-whisper covers about 100 languages without those turn events.

ElevenLabs' Scribe v2 Realtime is its live transcription model, at $0.39 an hour, and Scribe v2 handles recorded files at $0.22 an hour. For a phone agent, the realtime models are the ones that matter.

Open Source Alternative to Vapi / Retell

Self-hosted voice agent platform — no per-minute fees

dograh-hq/dograh

Star on GitHub

Agent platforms: Managed Agents vs ElevenAgents

The two agent prices cover different things. Cartesia lists Managed Agents at $0.03 a minute, down from $0.06, with $0.014 a minute on a Cartesia number and the language model on top, according to Cartesia's pricing page. ElevenAgents plans run from $6 to $990 a month with included minutes and $0.08 for each extra minute. Telephony is included, the language model is billed on top, and your phone provider bills its own carrier charges. Our Cartesia pricing breakdown works through Cartesia's side minute by minute.

ElevenAgents is the more developed platform today. It has visual Workflows, tests, keypad input and output, several transfer types and burst capacity above the plan's concurrency. Managed Agents launched on October 29 with system tools for keypad tones out, voicemail detection and transfer by number or SIP address, plus API batch calls. Its language models come from Cartesia's catalog, while ElevenAgents also takes your own OpenAI-compatible model.

Hosting is close. Both run in their own cloud, with residency and a business associate agreement (BAA) on Enterprise. ElevenLabs offers private deployment of ElevenAgents in your own AWS or GCP account, and Cartesia's sales team arranges on-premise or private-cloud deployment of its models. Our ElevenLabs review and ElevenLabs pricing posts cover that side in depth.

Who each one suits

Cartesia suits teams that want a top-ranked real-time voice on a credit plan, Ink 2's turn detection for English and the other four languages it covers, Hindi-English code-switching, and an agent minute priced before the model. Its numbers and agent tools are all in one place.

ElevenLabs suits teams that want the top-ranked voices, wider cloning tiers, an agent platform with workflows and tests already in place, and EU, India or Singapore residency on Enterprise. The surest way to choose is to run both voices through your own call script.

Using both inside Dograh

Dograh has Cartesia and ElevenLabs as built-in providers, each with your own key. In the voice settings you pick Sonic 3.6, 3.5 or 3, or ElevenLabs models from Flash v2.5 to Eleven v4 Turbo, and paste any voice ID, clones included. For ElevenLabs, the API address can point at a regional residency endpoint.

Voice and speech-to-text are separate settings, so one agent can speak with a Sonic voice and listen with Scribe v2 Realtime, or the other way round. In the transcriber settings, Ink 2 is offered for English and ink-whisper for other languages. Our Cartesia features post lists every Cartesia setting that carries over.

Mixing vendors means both see audio, since each still processes the speech it voices or transcribes, even on your own key. Self-hosted, Dograh adds no per-minute platform fee, so the bill is the two vendors' keys plus servers and phone minutes. Dograh Cloud adds its 1¢-a-minute platform fee on top.