ElevenLabs Quietly Made v4 Turbo the Default for Every New Voice Agent
Disclosure: this post contains an affiliate link to ElevenLabs. If you sign up through it, DailyAIPedia may earn a commission at no extra cost to you. ElevenLabs had.
Disclosure: this post contains an affiliate link to ElevenLabs. If you sign up through it, DailyAIPedia may earn a commission at no extra cost to you. ElevenLabs had no involvement in this post. Every item below was read off ElevenLabs’ own changelog on 12 October 2026.
On 5 October 2026 ElevenLabs changed the default model for conversational agents. Not added, not offered — changed. New agents now get eleven_v4_turbo unless you pick something else.
There was no blog post. ElevenLabs’ blog has published nothing since 30 September. This went out as a line in the changelog, which is a strange place to put a change that alters what everybody’s next agent sounds like.
If you follow this site you have already read about Eleven v4 itself, which launched on 28 September, and about ElevenAgents Architect, which arrived on 6 October. This is the piece between them, and it is the one that affects existing workflows rather than new toys.
| What changed | Detail | Who it affects |
|---|---|---|
| Default conversational model | Now eleven_v4_turbo | Anyone creating an agent from now on without explicitly choosing a model. |
| v4 and v4 Turbo in Conversational AI | Both now supported for conversational TTS | Agent builders who wanted v4 quality in a live conversation. |
| Two new agent LLMs | gpt-6.1-sol and claude-sonnet-5-5 | Anyone choosing the reasoning model behind an agent. |
| SDK support | Eleven v4 agent voice models in the JS and Python SDKs, v2.71.0 | Developers pinned to an older SDK version. |
| Realtime transcript editing | Speech-to-text transcript editing in client 1.26.0 and react 1.16.0 | Anyone correcting a live transcript mid-call. |
A changed default is not the same as a new option
Adding a model gives you a choice. Changing the default makes a choice for everyone who does not make one, which is most people most of the time.
Two practical consequences. If you build a new agent this week and do not touch the model selector, you are on v4 Turbo — so if it sounds different from the agent you built last month, that is why, and nothing is broken. And if you have been evaluating agents by spinning up fresh ones and comparing, your comparisons from before and after 5 October are not measuring the same thing.
What ElevenLabs does not publish is what the default was before, or any figure for how v4 Turbo compares with it on latency, quality or cost. We are not going to invent one. The honest position is that the default moved and you should listen to your own agent rather than trust ours or anyone else’s description.
Two frontier LLMs joined the agent brain list
Separately, ElevenLabs added gpt-6.1-sol and claude-sonnet-5-5 to the LLM options for agents. This is the model doing the thinking, as distinct from the model doing the speaking.
That distinction trips people up constantly, so to be explicit: an ElevenLabs agent has a voice model and a reasoning model, and they are configured separately. This release touched both ends on the same day — v4 Turbo on the voice side, two new options on the reasoning side — which is worth knowing if you are debugging an agent that got worse and are not sure which half to blame.
The bits a developer will actually feel
The rest of the 5 October release is API and SDK work, and some of it is genuinely useful:
preserve_voice_settingsfor transfers — voice configuration survives a transfer between agents, instead of resetting.exclude_transcriptfor post-call webhooks — stop sending the full transcript to your endpoint when you only need the metadata. A sensible privacy and payload-size win.wait_timeout_secson skip-turn tools — control how long an agent waits before moving on.- Client tool response timeouts accept
-1— an explicit no-timeout value. - Convai widget 0.18.3 — image previews and download links for attachments in agent messages.
- Merge proposals for agent branches, in alpha — review configuration changes and test results, comment, approve and merge. Team governance for agent configs.
One deprecation to note if you have an integration: structured procedures now compile and validate automatically when you update an agent or create a draft, so you no longer call the compile endpoint yourself, and an invalid procedure returns a 400. The Compile procedures endpoint is now legacy and ElevenLabs says new integrations should not use it.
Does v4 Turbo cost you more?
ElevenLabs states no price change in this release, and the plan prices we verified on 1 October — Starter at $6 a month for 30,000 credits, Creator at $22 for 121,000 — are unchanged.
There is one wrinkle worth knowing, and we are quoting it rather than paraphrasing because the wording is doing a lot of work. ElevenLabs’ pricing page, read on 1 October 2026, stated: “On Creator plans and above, up to 2x your monthly TTS credits used on v4 won’t count against your balance. Available in the web and mobile apps only.”
So v4 usage has been getting favourable credit treatment on Creator and above, in the apps but not via the API. What we cannot tell you is whether that still applies, or for how long — the same page carried a “Try v4 with 2x credits for two weeks” line, with no start date against it, which reads like a promotion rather than a standing term. If your usage depends on it, check your own credit balance after a session rather than assuming.
What we could not verify
Open questions, stated plainly
- What the previous default model was. ElevenLabs says the default changed to v4 Turbo but does not name what it replaced, so we cannot describe the before state.
- Whether existing agents were migrated. The changelog describes a changed default. It does not say whether agents created before 5 October stay on their original model or were moved.
- Any latency, quality or cost comparison. None published. “Turbo” implies faster; no figure supports it.
- How long the 2x v4 credit treatment lasts. The pricing page carried both a standing-sounding term and a two-week trial line, with no dates on either.
- Whether the new LLM options cost differently. Agent LLM choice has historically affected cost per conversation. Nothing is stated for gpt-6.1-sol or claude-sonnet-5-5.
All items above were read off ElevenLabs’ own changelog at elevenlabs.io/docs/changelog on 12 October 2026; the 5 October release is the only entry inside the first twelve days of the month. Plan prices and the v4 credit wording were verified against ElevenLabs’ own pricing page on 1 October 2026 as part of our monthly AI voice pricing check. If anything changes materially, this post gets a dated update box rather than a silent edit.
Frequently asked questions
What is the default ElevenLabs conversational model now?
Eleven v4 Turbo, identified as eleven_v4_turbo. ElevenLabs changed the default in its 5 October 2026 changelog release, so any agent you create without explicitly choosing a model now uses it.
Were my existing ElevenLabs agents moved to v4 Turbo?
ElevenLabs does not say. The changelog describes a changed default for new agents and says nothing about migrating existing ones. Check the model selector on an agent you built before 5 October rather than assuming either way.
Which LLMs can I use for an ElevenLabs agent?
The 5 October release added gpt-6.1-sol and claude-sonnet-5-5 to the existing options. Note that this is the reasoning model behind the agent, configured separately from the voice model that speaks.
Is Eleven v4 Turbo faster than the old default?
ElevenLabs publishes no latency, quality or cost comparison, and does not name the model v4 Turbo replaced. The name implies speed; nothing published measures it.
Does using v4 cost more ElevenLabs credits?
No price change was announced. ElevenLabs’ pricing page, read on 1 October 2026, stated that on Creator plans and above up to twice your monthly TTS credits used on v4 will not count against your balance, in the web and mobile apps only — not via the API. Whether that is permanent is unclear, since the same page also described a two-week v4 credit trial with no dates attached.
Do I need to update the ElevenLabs SDK?
For Eleven v4 agent voice models, yes — they arrived in the JS and Python SDKs at v2.71.0. Realtime speech-to-text transcript editing needs client 1.26.0 or react 1.16.0. Separately, if your integration calls the Compile procedures endpoint, that is now legacy: structured procedures compile automatically when you update an agent or create a draft.