Skip to content
Today
Sections

Build Reports disagree3 publishers 3 min read Published Updated

ElevenLabs aims v4 Turbo at live voice agents with a self-measured 150 ms to first speech

ElevenLabs released Eleven v4 Turbo for voice agents, reporting medians of about 100 ms inference latency and 150 ms to first speech. Moving an existing agent takes more than a model ID swap, since older voice clones need retraining and SSML break tags no longer work.

The Engineer · Build desk

Send a correction

What happened

  • Turbo adds bidirectional streaming, so developers can send text as a language model produces it and receive audio before the sentence is finished.
  • Artificial Analysis' Provider Voice Arena ranks the standard Eleven v4 first, with an Elo score of 1,319 from 1,674 samples.
  • ElevenLabs' X thread says an Instant Voice Clone can capture a voice from 10 seconds of audio, while its v4 documentation says a clone generally uses one to two minutes.
  • A two-week API launch offer prices Eleven v4 at $22 and Turbo at $11 per million characters.

Why it matters

  • Who has to decide decision Choosing a model for agent traffic means setting the arena-ranked standard v4 against Turbo's lower latency and half-price launch rate, and only a listening test on a team's own voices settles it.
  • What it limits constraint A multilingual agent that must keep one speaker's accent across languages starts from the wrong default, because v4 steers cross-language cloned speech toward a fluent target-language accent.
  • What doesn't add up contradiction Runtimewire says v4 comes at no extra cost on eligible Creator+ plans within credit limits, while TestingCatalog says both models are on every plan including the free tier, so the cost of a small pilot depends on which report holds.

Turbo comes with two latency figures, and they measure different points in a request. The roughly 100 ms figure is ElevenLabs' reported median inference latency [2]. The 150 ms figure is the median time to first speech [3]. ElevenLabs is positioning Turbo for support, sales and scheduling agents [12]. On a call, time to first speech is closer to the pause a caller hears. It is also the figure ElevenLabs used against competitors [4]. Both are medians. Neither report includes tail percentiles, and an agent's slowest turns sit in the tail.

Reaching that figure inside an agent loop depends on the wiring. ElevenLabs offers streaming and non-streaming endpoints [19]. Bidirectional streaming on Turbo [13] lets synthesis start while the language model is still writing. An agent that buffers the full LLM reply before calling synthesis puts the reply's generation time in front of Turbo's 150 ms.

ElevenLabs' comparison puts Turbo 112 ms ahead of Cartesia Sonic 3.6 and 664 ms ahead of OpenAI's GPT-4o mini TTS [20]. ElevenLabs ran that test [4]. For the 112 ms lead to hold on production calls, a team's network path to each provider, its text chunk sizes and its voices would need to resemble what ElevenLabs measured. A 664 ms gap leaves room for setup differences. The 112 ms gap is the one I would re-measure on my own traffic before switching.

The quality evidence belongs to the other model. The arena ranking is for the standard Eleven v4 [5], and the arena scores each provider's native voices [6]. A cloned voice on Turbo, delivered as telephony-ready mu-law audio [18], is outside what it measured. Runtimewire reports that ElevenLabs' post cites the ranking without spelling out the methodology [7].

Swapping the model ID is a one-line change [16]. Retraining the older clones [14] is where ElevenLabs' two cloning figures collide. The documentation's usual Instant Voice Clone sample is six to twelve times the length the launch thread promised [27]. Both numbers come from ElevenLabs, about the same feature. I'd retrain against the documentation and treat the 10 seconds as a best case [25]. Professional Voice Clones, unavailable in v3, are back in v4 [9], and Turbo keeps the same Professional Voice Clone across a call [17].

Prompt code needs its own change. v4 disables SSML break tags and takes pacing from natural-language audio tags, the bracketed directions such as laughs and whispers that go straight into the script [15]. An agent that inserts `<break>` tags to space out a confirmation number has to move that pacing into audio tags.

At launch rates, a maximum-length generation of 10,000 characters [10] costs $0.22 on v4 and $0.11 on Turbo [22]. The offer runs for two weeks [11].

What to watch

  • An independent measurement of Turbo's time to first speech that reports p95 and p99 latency alongside the median.
  • The per-character rates for Eleven v4 and Turbo once the two-week API launch offer ends.
  • Whether ElevenLabs revises the 10-second cloning claim or the one-to-two-minute sample guidance in its v4 documentation.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence50
Adoption
Not scored
Hype gap+25
Incentives70
Confidence60

Perspective Coverage

3 publishers
Builder
Builder 45%
Operator
Operator 33%
Investor
Investor 22%
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    ElevenLabs released Eleven v4 and Eleven v4 Turbo, announced in a September 28 post on X, available through ElevenAgents, ElevenCreative and ElevenAPI.

    Reported SupportedView cited source
  2. [2]

    ElevenLabs says v4 Turbo has a median inference latency of about 100 milliseconds, a company-reported figure.

    Reported Supported Source: ElevenLabs, as reported by Runtimewire and TestingCatalogView cited source
  3. [3]

    Eleven v4 Turbo has about 150 ms median time to first speech.

    Reported Supported Source: ElevenLabs, as reported by TestingCatalogView cited source
  4. [4]

    In ElevenLabs' comparison, Turbo reached speech in 150 ms, versus 262 ms for Cartesia Sonic 3.6 and 814 ms for OpenAI GPT-4o mini TTS.

    Reported Supported Source: ElevenLabs' own comparison, as reported by TestingCatalogView cited source
  5. [5]

    Artificial Analysis' Provider Voice Arena leaderboard lists Eleven v4 at No. 1, with an Elo score of 1,319 from 1,674 samples.

    Reported SupportedView cited source
  6. [6]

    The Provider Voice Arena leaderboard compares models using each provider's native voices.

    Reported SupportedView cited source
  7. [7]

    ElevenLabs' post cites the leaderboard ranking without spelling out the methodology.

    Reported Supported Source: RuntimewireView cited source
  8. [8]

    According to the v4 documentation, when generated speech is in a different language from the source recording, the model aims for a fluent target-language accent rather than retaining the source speaker's accent by default.

    Reported SupportedView cited source
  9. [9]

    Professional Voice Clones return in v4 after being unavailable in v3.

    Reported SupportedView cited source
  10. [10]

    Each v4 generation supports up to 10,000 characters.

    Reported SupportedView cited source
  11. [11]

    The API launch offer runs for two weeks: Eleven v4 at $22 per million characters and Turbo at $11 per million characters.

    Reported SupportedView cited source
  12. [12]

    ElevenLabs is positioning v4 Turbo for support, sales and scheduling agents.

    Reported SupportedView cited source
  13. [13]

    Eleven v4 Turbo adds bidirectional streaming, allowing developers to send text as a language model produces it and receive audio before a sentence is finished.

    Reported SupportedView cited source
  14. [14]

    All 17,500-plus voices in ElevenLabs' library work with v4, though older Instant and Professional Voice Clones need retraining.

    Reported SupportedView cited source
  15. [15]

    SSML break tags are disabled in v4, so pauses and delivery are controlled through natural-language audio tags such as [laughs] and [whispers] placed directly in a script.

    Reported SupportedView cited source
  16. [16]

    Developers can switch models with a single model ID.

    Reported SupportedView cited source
  17. [17]

    Eleven v4 Turbo retains v4's expressive controls and supports the same Professional Voice Clone across a call.

    Reported SupportedView cited source
  18. [18]

    ElevenLabs output formats include MP3, WAV, PCM and telephony-ready mu-law.

    Reported SupportedView cited source
  19. [19]

    Developers can use streaming or non-streaming endpoints through REST, TypeScript and Python tools.

    Reported SupportedView cited source
  20. [20]

    In ElevenLabs' comparison, Turbo is 112 ms faster to first speech than Cartesia Sonic 3.6 and 664 ms faster than OpenAI GPT-4o mini TTS.

    Derived Supported
  21. [21]

    Turbo's launch rate is half of Eleven v4's launch rate per million characters.

    Derived Supported
  22. [22]

    At launch rates, a maximum-length 10,000-character generation costs $0.22 on Eleven v4 and $0.11 on Turbo.

    Derived Supported
  23. [23]

    ElevenLabs says v4 is included at no additional cost for eligible Creator+ plans, subject to monthly credit limits.

    Reported Contested Source: RuntimewireView cited source
  24. [24]

    Both models are included on every plan, including a free tier with 10,000 monthly credits, while paid plans start at $6 per month.

    Reported Contested Source: TestingCatalogView cited source
  25. [25]

    ElevenLabs' X thread says Instant Voice Clones can capture a voice from 10 seconds of audio.

    Reported Contested Source: ElevenLabs X thread, as reported by RuntimewireView cited source
  26. [26]

    ElevenLabs' v4 documentation says an Instant Voice Clone generally uses a sample of one to two minutes.

    Reported Contested Source: ElevenLabs v4 documentation, as reported by RuntimewireView cited source
  27. [27]

    The documented Instant Voice Clone sample of one to two minutes is six to twelve times the 10 seconds claimed in the launch thread.

    Derived Contested

Sources

3 independent publishers whose own reporting we read for this story.

  1. runtimewire.com

    1 article · September 28, 2026

    ElevenLabs ships v4 voice models, ranked No. 1 by Artificial Analysis
  2. testingcatalog.com

    1 article · September 28, 2026

    ElevenLabs launches Eleven v4 and v4 Turbo voice models
  3. the-decoder.com

    1 article · September 29, 2026

    ElevenLabs' new v4 speech model makes AI voices more expressive and consistent

Topics and entities

Topics

Entities