Realtime models | ElevenLabs Documentation

Overview

A Realtime model is an ElevenLabs model built for low-latency voice conversation, available to voice agents.

Every voice agent has a Voice mode: Custom or Realtime. In Custom mode, you choose the LLM, TTS model, and ASR settings yourself, as described in Models. In Realtime mode, you choose a Realtime model, and ElevenLabs manages those settings for you.

Two Realtime models are available:

Model	`realtime_model` value	Description
Eleven Realtime v1 Mini	`eleven_realtime_v1_mini`	The default Realtime model. Lowest latency and cost.
Eleven Realtime v1 Max	`eleven_realtime_v1_max`	Reasons before it responds, for complex instructions and multi-step tool use. Costs more than Mini.

The agent’s system prompt, voice, tools, knowledge base, and turn settings work the same in both modes.

Coming from another realtime API, such as SpaceXAI or OpenAI? The Realtime API is a quicker way to try out Realtime models: point your existing client at ElevenLabs and keep the events you already use.

Realtime models are rolling out gradually. If the Voice mode setting doesn’t appear on your agent’s Agent tab, Realtime models aren’t available for your workspace yet, and the API rejects realtime_model with the error Realtime models are not available for this workspace.

Choosing a voice mode

Keep the agent in Custom mode if any of the following apply:

The agent uses a custom LLM, Amazon Bedrock, or Speech Engine.
The agent needs a specific LLM or TTS model.
Workflow nodes override the LLM, TTS model, or another managed setting.
The agent transfers to or from agents in Custom mode.
The agent is text-only. Realtime models are built for voice conversations.

Otherwise, start with Eleven Realtime v1 Mini. Choose Eleven Realtime v1 Max instead when:

The agent follows long or complex instructions, or relies on multi-step tool calls, and Mini’s responses aren’t good enough.
The agent must comply with Zero Retention Mode or HIPAA in the US or EU. Mini isn’t available under either.

Max isn’t available in the India and Singapore data residency environments, and it needs a token limit of at least 1,024, or no limit. See Limitations and Compliance and data residency.

Configuration

Update via the dashboard

Update via the CLI

Update via the API

Open the Agent tab

Open your agent in the dashboard and go to the Agent tab.

Set the voice mode

Set Voice mode to Realtime. The agent switches to Eleven Realtime v1 Mini.

Choose a model

To use Eleven Realtime v1 Max instead, select it from the Model list.

Publish

Publish your changes. Publishing fails if the agent breaks one of the rules in Limitations.

To turn Realtime off, set Voice mode to Custom and publish. The agent goes back to the LLM, TTS model, and ASR settings stored on it.

Managed settings

While a Realtime model is set, ElevenLabs manages the following settings, and the agent ignores the values stored for them. The stored values come back into use when you switch to Custom.

Setting	Field in `conversation_config`
LLM	`agent.prompt.llm`
Reasoning effort	`agent.prompt.reasoning_effort`
Thinking budget	`agent.prompt.thinking_budget`
Parallel tool calls	`agent.prompt.enable_parallel_tool_calls`
Cascade timeout	`agent.prompt.cascade_timeout_seconds`
TTS model	`tts.model_id`
Expressive mode	`tts.expressive_mode`
ASR provider	`asr.provider`

The token limit (agent.prompt.max_tokens) keeps its stored value, but workflow nodes can’t override it, and Eleven Realtime v1 Max needs a minimum value. See Limitations.

To change any of these settings, switch the agent to Custom.

Settings that still apply

All other agent settings work as they do in Custom mode, including:

the system prompt and first message
the voice, including multi-voice support
languages
temperature
tools and the knowledge base
conversation flow settings such as turn eagerness, timeouts, and interruptions

The dashboard hides the LLM settings, including temperature, while the agent is in Realtime mode. Use the CLI or the API to change temperature on a Realtime agent.

If the Realtime model doesn’t respond in time or returns an error, the agent falls back to backup LLMs as described in LLM cascading. The Realtime model sets how long the agent waits before falling back.

Limitations

ElevenLabs checks these rules when you publish the agent. A draft that breaks one of them still saves, but publishing fails with an error that names the setting to change.

ElevenAgents agents only: Agents created through other ElevenLabs products, such as Speech Engine, can’t use a Realtime model.
No custom LLM: Realtime mode isn’t available while the agent’s LLM is a custom LLM or Amazon Bedrock. Select a built-in LLM in Custom mode before switching.
No workflow overrides of managed settings: A workflow node can’t override any managed setting or the token limit. Remove the override, or keep the agent in Custom mode.
Token limit for Max: Eleven Realtime v1 Max needs a token limit of 1,024 or more, or no limit.
Transfers between modes: An agent in Realtime mode can only transfer to and from other agents in Realtime mode. Transfers between Mini and Max agents are supported.

Publishing checks only the transfers the published agent makes. A transfer into a Realtime agent from a Custom-mode agent is caught only when the Custom-mode agent is next published; until then, the transfer is rejected during the conversation and the current agent stays on the call.

Transfer errors read Transfers between default and Realtime model modes are not supported., where “default” means Custom mode.

Compliance and data residency

Realtime models follow the same Zero Retention Mode, HIPAA, and data residency rules as other models. Availability differs between Mini and Max:

Environment	Eleven Realtime v1 Mini	Eleven Realtime v1 Max
US	Available	Available
EU data residency	Available	Available
India data residency	Available	Not available
Singapore data residency	Available	Not available
Zero Retention Mode and HIPAA	Not available	Available in the US and EU

Zero Retention Mode applies whether it is enforced on the workspace or enabled on the agent.

If a Realtime model isn’t available under the agent’s compliance or data residency settings, ElevenLabs rejects it with an error that names the model. Choose the other Realtime model, or switch to Custom.

Pricing

A Realtime agent is billed the standard ElevenAgents per-minute rate, plus its input, cached input, and output tokens at the Realtime model’s rates. Eleven Realtime v1 Max costs more per token than Eleven Realtime v1 Mini. See ElevenLabs pricing for current rates.