Models

ElevenAgents provides a unified interface to connect your agent to multiple models and providers, offering flexibility, reliability, and cost optimization.

Key features

• Unified access: Switch between providers and models with minimal code changes
• High reliability: Automatically cascade from one provider to another if one fails
• Spend monitoring: Monitor your spending across different models

Supported models

Currently, the following models are natively supported and can be configured via the agent settings:

Custom LLM

Using your own custom LLM is supported by specifying the endpoint we should make requests to and providing credentials through our secure secret storage. Learn more about custom LLM integration.

Choosing a model

Selecting the most suitable LLM for your application involves considering several factors:

• Task complexity: More demanding or nuanced tasks generally benefit from more powerful models (e.g., OpenAI’s GPT-4 series, Anthropic’s Claude Sonnet 4, Google’s Gemini 2.5 models)
• Latency requirements: For applications requiring real-time or near real-time responses, such as live voice conversations, models optimized for speed are preferable (e.g., Google’s Gemini Flash series, Anthropic’s Claude Haiku, OpenAI’s GPT-4o-mini)
• Context window size: If your application needs to process, understand, or recall information from long conversations or extensive documents, select models with larger context windows
• Cost-effectiveness: Balance the desired performance and features against your budget. LLM prices can vary significantly, so analyze the pricing structure (input, output, and cache tokens) in relation to your expected usage patterns
• HIPAA compliance: If your application involves Protected Health Information (PHI), it is crucial to use an LLM that is designated as HIPAA compliant and ensure your entire data handling process meets regulatory standards

Model configuration

Temperature

Temperature controls the randomness of model responses. Lower values produce more consistent, focused outputs while higher values increase creativity and variation.

• Low (0.0-0.3): Deterministic, consistent responses for structured interactions
• Medium (0.4-0.7): Balanced creativity and consistency
• High (0.8-1.0): Creative, varied responses for dynamic conversations

Backup LLM configuration

Configure backup LLMs to ensure conversation continuity when the primary LLM fails or becomes unavailable.

Configuration options:

• Default: Uses ElevenLabs’ recommended fallback sequence
• Custom: Define your own cascading sequence of backup models
• Disabled: No fallback (strongly discouraged for production)

Learn more about LLM cascading.

Thinking budget

Control how many internal reasoning tokens the model can use before responding. More tokens improve answer quality but slow down response time.

Options:

• Disabled: Fastest replies with no internal reasoning overhead
• Low: Minimal reasoning for quick responses
• Medium: Balanced reasoning and speed
• High: Maximum reasoning for complex queries

Reasoning effort

Some models support configurable reasoning effort levels (None, Low, Medium, High).

For conversational use-cases:

Keep reasoning effort set to None to avoid the agent thinking too long, which can disrupt natural conversation flow.

For workflow steps:

Reasoning effort is perfect for workflow steps that require complex thought or decision-making where response time is less critical.

Reasoning summary

Reasoning summary stores a summary of the model’s reasoning on the conversation transcript, so you can review why the agent responded or called a tool the way it did. Messages that involved reasoning show a Reasoning badge in the conversation history; select it to expand the summary.

Enable the Reasoning summary toggle (off by default) in the agent’s LLM settings, or set enable_reasoning_summary via the API.

Limitations:

• Summaries only appear for turns where the model actually reasoned (see Reasoning effort). The setting stores the summary; it does not change how the model reasons.
• Some providers respond faster when no summary is requested, so test the latency impact before enabling it on latency-sensitive voice agents.
• Providers decide whether to return a summary, so some reasoned turns will have none. ElevenLabs-hosted models return the full reasoning trace.
• Summaries are part of the stored transcript: they are not streamed during the conversation, follow retention and PII redaction settings, and are unavailable with Zero Retention Mode.

Summaries are also returned in the reasoning field of each transcript item in the Get conversation response.

Understanding pricing

• Tokens: LLM usage is typically billed based on the number of tokens processed. As a general guideline for English text, 100 tokens is approximately equivalent to 75 words
• Input vs. output pricing: Providers often differentiate pricing for input tokens (the data you send to the model) and output tokens (the data the model generates in response)
• Cache pricing:
  • input_cache_read: This refers to the cost associated with retrieving previously processed input data from a cache. Utilizing cached data can lead to cost savings if identical inputs are processed multiple times
  • input_cache_write: This is the cost associated with storing input data into a cache. Some LLM providers may charge for this operation
• The prices listed in this document are per 1 million tokens and are based on the information available at the time of writing. These prices are subject to change by the LLM providers

For the most accurate and current information on model capabilities, pricing, and terms of service, always consult the official documentation from the respective LLM providers (OpenAI, Google, Anthropic).

HIPAA compliance

Certain LLMs available on our platform may be suitable for use in environments requiring HIPAA compliance, please see the HIPAA compliance docs for more details.

Related resources

• Custom LLM integration
• LLM cascading
• Optimizing costs

Models Hosted by ElevenLabs

ElevenLabs offers access to a variety of AI models, including select third-party models that are hosted by ElevenLabs.

When a model is designated as “Hosted by ElevenLabs”, the model is deployed and operated on infrastructure managed by ElevenLabs. These models are currently hosted in the United States, or in the region of your enterprise deployment.

How to identify models Hosted by ElevenLabs

Models hosted by ElevenLabs are clearly labeled in the ElevenLabs interface. Look for the “Hosted by ElevenLabs” designation when browsing or selecting models.

How your data is handled

For models hosted by ElevenLabs, requests are processed within ElevenLabs-managed infrastructure. This means that the original model provider does not receive, access, or process the inputs and outputs submitted through ElevenLabs.

By hosting these models, ElevenLabs can provide a consistent experience while maintaining control over the infrastructure used to serve model requests.

Frequently asked questions

Where are self-hosted models hosted?

Models hosted by ElevenLabs are currently hosted on infrastructure located in the United States, or in the region of your enterprise deployment.

Does the original model provider access my data or metadata of my usage?

For models hosted by ElevenLabs, the model’s original developer never receives your inputs or outputs and has no access to the data about the deployment that processes them.

How do I know whether a model is Hosted by ElevenLabs?

Models hosted by ElevenLabs are clearly identified in the ElevenLabs interface with a “Hosted by ElevenLabs” designation.