Learn
ElevenLabs v4 vs v3: Languages, Length Limits, Latency and Price
Eleven v4 costs the same as v3, $0.08 per 1,000 characters on the API, and runs on every ElevenLabs plan including Free. It adds 90+ languages against 70+, 10,000 characters per request against 5,000, and a v4 Turbo at about 100 ms.
UpdatedOctober 23, 2026
In short
Use Eleven v4 for new ElevenLabs work: it costs the same as Eleven v3, $0.08 per 1,000 characters on the API, and runs on every subscription plan including Free with no separate credit rate. Against v3 it covers 90+ languages instead of 70+, takes 10,000 characters per request instead of 5,000, and comes with Eleven v4 Turbo for voice agents at about 100 ms median inference latency and $0.04 per 1,000 characters. Re-test voices you tuned on v3 before switching a long-running project, because the speaker-identity method is new.
On this page3
Eleven v4 and v3 side by side
Model | Languages | Characters per request | Median latency | API price per 1,000 characters |
|---|---|---|---|---|
Eleven v4 | 90+ | 10,000 | Not published | $0.08 |
Eleven v4 Turbo | 90+ | Not published | About 100 ms | $0.04 |
Eleven v3 | 70+ | 5,000 | Not published | $0.08 |
Eleven v3 Conversational | 70+ | Not published | About 280 ms | — |
Latency figures are ElevenLabs' and exclude application and network time; ToolColumn did not check v3 Conversational's API price. Read the table in pairs: v4 against v3 for narration and dialogue, v4 Turbo against v3 Conversational for real-time agents. In both pairs the v4 model covers more languages and, where ElevenLabs publishes it, does more per request or answers faster.
Twice the text per request, 90+ languages and 10-second clones
Longer scripts in one request. At 10,000 characters, v4 takes about twice as much text per call as v3, roughly 1,500 English words against 750 by ToolColumn's estimate of about six and a half characters per word. A narrator splitting chapters for v3 can send chunks twice the size, which means fewer joins to listen for.
More languages, steadier accents. ElevenLabs lists 90+ languages for v4 against 70+ for v3; its model page names each v4 language. ElevenLabs says a voice recorded in one language now speaks the others with a native accent and no longer drifts back toward its source accent during a long generation.
Clones from 10 seconds of audio. Instant Voice Clones on v4 capture a voice from about 10 seconds of audio, and ElevenLabs reports closer similarity to the source voice on both v4 models.
Direction you can write. Both v4 models follow inline audio tags such as [laughs] or [phone buzzing] and plain-language delivery notes more closely than earlier models, according to ElevenLabs, and IPA phoneme support for custom pronunciations is improved.
Quality claims to read as claims. ElevenLabs says Eleven v4 ranks first on Artificial Analysis and that about 75% of listeners preferred it in blind head-to-head tests against competing models. Both figures come from ElevenLabs' launch post, not from an independent test ToolColumn ran.
The same $0.08 on the API, and v4 on every plan
The switch costs nothing extra. On the API, v4 and v3 share the $0.08 rate, and v4 Turbo halves it for agents. On subscription plans, text to speech uses 1 credit per character and ElevenLabs' plan comparison marks Eleven v4 on every plan without listing a different rate, so ToolColumn expects a plan's credits to buy as much v4 speech as they bought v3 until ElevenLabs states otherwise. ElevenLabs offers v4 in ElevenAgents, ElevenCreative and the API, where it runs through the Text to Dialogue endpoint; v3 remains in the model list for projects that need it.
Official sources
Editorial guidance grounded in official product sources.
FAQ
Common questions
Is ElevenLabs v4 more expensive than v3?
No. On the API, Eleven v4 costs $0.08 per 1,000 characters, the same as Eleven v3 and Multilingual v2, and Eleven v4 Turbo costs $0.04. On subscription plans ElevenLabs lists no separate credit rate for v4.
Is Eleven v4 available on the ElevenLabs free plan?
Yes. ElevenLabs' plan comparison marks the Eleven v4 model on every plan, Free included, and the launch post invites free accounts to start generating with it.
How much text can Eleven v4 read in one request?
Up to 10,000 characters, twice Eleven v3's 5,000. Multilingual v2 also takes 10,000 and Flash v2.5 takes 40,000.
What is the difference between Eleven v4 and Eleven v4 Turbo?
Both use the new v4 architecture and support 90+ languages. v4 is the expressive model for narration, dialogue and audiobooks; v4 Turbo is the low-latency version for voice agents, with a median inference latency of about 100 ms and half the API price.
Next steps