Eleven v4 Turbo: ~150ms First Speech and WebSocket Integration

ElevenLabs officially made Eleven v4 and its low-latency Eleven v4 Turbo variant available on September 28, 2026. A developer X post dated October 21 was a later communication about the model, not its initial launch.

Image source: ElevenLabs Developers (@ElevenLabsDevs)

Eleven v4 Turbo targets responsive conversational voice agents. ElevenLabs reports a median time to first speech of approximately 150ms and a separate median inference latency of approximately 100ms. These are different metrics, not universal per-request guarantees.

~150ms Time to First Speech and the Documented Integration Path

The model focuses on reducing the delay before speech begins in interactive applications.

ElevenLabs' product page cites a median time to first speech of approximately 150ms. Actual latency depends on the request and deployment environment.

• Standard Eleven v4: ElevenLabs documents eleven_v4 via the Text to Dialogue API for content creation and long-form audio.
• Eleven v4 Turbo: The documented path for eleven_v4_turbo is the Text to Dialogue WebSocket for agents and interactive applications.

What to Check Before Migrating from Eleven v4

The documented interfaces differ, so changing only a model identifier is not a guaranteed drop-in migration.

• Endpoint and protocol: Compare Text to Dialogue API and WebSocket connection, request, and response requirements against current documentation.
• Streaming behavior: Test initial audio latency, chunk handling, errors, and reconnect behavior in the actual application.
• SDK compatibility: Verify support for the relevant WebSocket features in the specific SDK language and version being used.

Practical Considerations for Voice Agent Pipelines

Eleven v4 Turbo is positioned directly for conversational infrastructure where response latency directly affects conversational flow.

ElevenLabs positions standard Eleven v4 for content creation and long-form audio, and Turbo for real-time agents and interactive use. Teams should test both latency and output quality for their specific workload.

Treat a migration as a documented WebSocket integration change rather than assuming a shared API format or a one-parameter switch.

Sources

• ElevenLabs official changelog: September 28, 2026 availability and API routes
• ElevenLabs product page: Eleven v4 Turbo performance metrics
• ElevenLabs Developers official X (@ElevenLabsDevs): October 21 follow-up