Eleven v4 and Turbo split narration from live voice work

A finished voiceover and a live voice agent have different failure points. A voiceover can be regenerated before export; a conversation has to respond while someone waits. ElevenLabs’ September 28, 2026 release introduces Eleven v4 and v4 Turbo for those two jobs. Its launch post, revised October 1, lists availability in ElevenCreative, ElevenAgents and ElevenAPI. We checked the sources on October 4. ElevenLabs announcement.

Pick the path that matches the deliverable

The developer changelog specifies eleven_v4 through Text to Dialogue for content and long-form audio, and eleven_v4_turbo through the Text to Dialogue WebSocket for interactive work. That is more useful migration information than assuming an older text-to-speech integration can keep every request unchanged. September 28 developer release notes.

ElevenLabs reports about 100ms median inference latency for Turbo, while its launch article reports about 150ms to first audible speech under a test that removes network latency. Those measure different stages. Neither is a measured end-to-end response time for your agent; the rest of its processing and network still matter. Release notes, launch methodology.

For an edited social ad, our recommendation is to start with delivery rather than the fastest response. An illustrative trial would use the same short script and voice in your current workflow and v4. Include the product name, a quiet opening and a firm closing line. Listen for accurate pronunciation, natural emphasis and consistent identity between lines. Save rejected takes and the reason for rejection. This is a proposed test, not a report of our listening results.

Two migration details deserve attention

The model documentation lists Stability and Similarity as voice settings; Style and Speed sliders and SSML are unavailable. It also describes changed cross-language accent handling: when the target language differs from the reference voice’s language, v4 adopts the target language’s native accent. A workflow that deliberately retains the speaker’s original accent needs a direct comparison before switching. Eleven v4 controls and accent documentation.

That distinction can matter more than general quality praise. For example, a narrator’s characteristic accent may be part of the creative brief. A technically clear translation that changes that identity can still miss the intended result.

The checked Creative pricing page separates Free from paid tiers and lists a Commercial License under Starter. We have not established a model-specific v4 credit multiplier or a complete API job price. Do not convert a plan’s credit allowance into guaranteed v4 minutes. ElevenLabs pricing.

Next action: compare one short, rights-cleared script before moving a production project. Developers should verify the documented endpoint and supported settings; creators should compare usable takes and correction time. Continue with the ElevenLabs guide and the podcast voice workflow for the larger production decision.