ElevenLabs in Australia: pricing, latency and data residency

ElevenLabs models for voice agents

ElevenLabs lists seven text to speech models and three speech to text models in its docs. Flash v2.5 and v4 Turbo are its two low-latency models for phone agents.

The latency figures in the table are ElevenLabs' own claims, measured at its servers. The time a caller in Australia waits is longer, because the audio travels from Singapore or the United States.

ElevenLabs pricing plans

ElevenLabs sells monthly plans in US dollars. Each plan carries a pool of credits, and text to speech draws credits by the character. The agent minutes in the table come out of the same pool.

The Free plan gives 10,000 credits a month, enough for about 11 minutes of speech. Commercial use starts on Starter.

Cost per spoken minute

A caller hears about 150 words a minute, which is close to 900 characters of text. At that rate, Flash v2.5 costs about USD 0.036 per minute of agent speech and v4 costs about USD 0.072.

Scribe v2 Realtime adds USD 0.0065 per minute of caller audio. The ElevenLabs agent platform bundles both at USD 0.08 a minute, and bills the language model on top.

ElevenLabs does not charge for the phone line. Your carrier, such as Twilio or a SIP provider, bills the phone line directly.

Where ElevenLabs processes Australian data

The default API address, api.elevenlabs.io, is not pinned to a country. When we connected from Sydney, the streaming WebSocket answered from Google Cloud asia-southeast1 in Singapore in 3 of 3 tries, and the REST API answered from us-central1 in Iowa in 6 of 6.

ElevenLabs sells data residency on Enterprise only, in three isolated regions. Each region has its own API address, and none is in Australia.

• European Union: api.eu.residency.elevenlabs.io
• India: api.in.residency.elevenlabs.io
• Singapore: api.sg.residency.elevenlabs.io

Limits of ElevenLabs residency

ElevenLabs' residency docs say support staff and content moderation may still process data outside the chosen region. Residency does not turn on zero retention, which is a separate Enterprise setting.

Zero Retention Mode covers text to speech, speech to text and agents. It does not cover voice cloning, dubbing, music or Studio, so a cloned voice keeps its training audio on ElevenLabs' servers.

ElevenLabs latency measured from Sydney

We measure time to first audio on every turn of every live call: the gap between sending text to the voice provider and receiving the first audio back. Our call servers run in Sydney.

Over the 14 days to 4 October 2026, ElevenLabs took 499 ms at the median and 842 ms at the 90th percentile. Cartesia, on the same servers in the same window, took 448 ms and 604 ms.

An earlier 30-day read to 22 August 2026 compared ElevenLabs models. Flash v2.5 and Turbo v2.5 differed by 95 ms at the median, less than the 125 ms gap to Cartesia in that window. Distance to the server explains part of the gap and the model explains the rest.

Running ElevenLabs in your own cloud

ElevenLabs sells a private deployment into your own AWS or Google Cloud account. It covers the agent platform, the v2 and v2.5 voice models and Scribe v2, and ElevenLabs says all text, audio and call data stays inside your infrastructure. Azure support is listed for the second half of 2026.

The voice and transcription models can also run alone as endpoints on Amazon SageMaker or Google Vertex AI. The full deployment docs sit behind an NDA, so region support, including Sydney, has to come from ElevenLabs sales.

On-premises and on-device deployment was announced as early access on 9 April 2026. We found no general availability date on elevenlabs.io.

ElevenLabs compliance and contracts

ElevenLabs states SOC 2 Type II, ISO 27001 and PCI DSS Level 1 certification on its trust page, with HIPAA and GDPR attestations. These apply to the company, not to a plan.

A HIPAA business associate agreement needs Enterprise with Zero Retention Mode switched on. Custom single sign-on and static SIP addresses are Enterprise features too.

Connecting ElevenLabs to a phone line

ElevenLabs agents answer phone calls through a native Twilio integration or a SIP trunk at sip.rtc.elevenlabs.io. The SIP trunk takes G.722 and G.711 audio, and ElevenLabs names Telnyx, Vonage, Sinch, Plivo and Bandwidth among the carriers that work with it.

A team building its own agent can stream speech from the text to speech WebSocket instead, and keep the call on its own carrier and its own servers.

Where ElevenLabs fits an Australian deployment

ElevenLabs has the widest range of voices and languages of the three providers we run. Our voice agents use an ElevenLabs voice by default, and we keep it where the voice matters more than 50 ms of latency.

If your compliance team needs the voice processed in Australia, ElevenLabs cannot do it on its public API today. Deepgram runs text to speech in Sydney on any plan, and Cartesia sells an Australian deployment on Enterprise.