On September 28, 2026, ElevenLabs shipped its biggest voice-model release of the year: Eleven v4 and its real-time sibling Eleven v4 Turbo. Two weeks later, the launch is still the talk of the voice-AI world — and if you use AI voices for anything from YouTube narration to voice agents, the details matter more than the headlines.
As always on this site: this is based on ElevenLabs’ launch announcements and verified reporting as of October 12, 2026. I haven’t tested the models hands-on myself, so this is a reporter’s explanation, not a lab review.
Two models, two jobs
ElevenLabs released two models under the IDs eleven_v4 and eleven_v4_turbo, and they’re designed for different work:
Eleven v4 is the flagship for anything a listener hears later: audiobooks, character voiceovers, dialogue, dubbing. ElevenLabs calls it the company’s highest quality model, promising “reliable speaker preservation across long text generations” — in plain terms, the voice shouldn’t drift into someone else’s halfway through a long chapter. One request can now carry 10,000 characters, roughly ten minutes of audio; v3 capped out at half that, which is a genuine workflow upgrade for long-form narration.
Eleven v4 Turbo is the real-time sibling, aimed at voice agents, assistants, and interactive characters. The Turbo documentation quotes a median inference latency of about 100 milliseconds, and the launch post adds a median time to first speech of about 150 ms — with the bold claim that Turbo can answer faster than the average pause between two people talking. An independent writeup from the Micdrop project measured roughly 90 ms to start speaking a short sentence on Turbo versus about 600 ms on the standard model, which lines up with the “real-time vs. narration” split.
The headline features
90+ languages, up from 70+ on v3. The jump matters less as a number and more in the details: TechCrunch reported the biggest quality gains showed up in Japanese, Brazilian Portuguese, Mandarin, and Cantonese — exactly the languages where AI voices have historically sounded the most robotic.
Expression control through audio tags. This is the change that has voice designers excited. You can now write directions directly into your script — tags like [laughs] or [whispers] — and the model acts them out in sentence. Reporting suggests v4 follows these tags far more consistently than most TTS engines have historically honored their direction systems. One practical consequence: if your existing scripts lean on SSML prosody tags for pacing and emphasis, you’ll need to rewrite that direction as natural-language audio tags instead. More work up front, but the results reportedly track your intent more reliably.
Better cloning. Professional Voice Clones are fully supported — the documentation admits they weren’t fully optimized on v3 — and ElevenLabs says an Instant Voice Clone can reach high fidelity from just 10 seconds of audio. That’s the vendor’s claim, and I’d want to hear the results before repeating it as fact, but the direction is clear.
The benchmark claims — read carefully
ElevenLabs says v4 took first place on the Artificial Analysis voice arena and claims roughly 75% listener preference in blind tests against Cartesia, Inworld, and two Gemini TTS models. These are real data points, but they’re vendor-supplied, and the arena tests specific conditions that may or may not match your use case. I’d file them under “promising, verify with your own ears” rather than “settled fact.” The language-coverage numbers and latency figures are more concrete because multiple independent writeups confirm them.
Pricing: what v4 actually costs
List pricing on the API: v4 at $0.08 per 1,000 characters and v4 Turbo at $0.04 per 1,000 characters. But there’s a catch that makes today significant: ElevenLabs has been running a 72% launch discount — until October 12, 2026, i.e., today. Under the discount, v4 costs $0.022 per 1,000 characters and Turbo $0.011, with the struck-through list rates shown on the pricing page as what comes after.
So treat the discount as reportedly ending today and check the live pricing page before you commit — launch-week details change fast, and there’s no guarantee ElevenLabs won’t extend it. For production planning, the safer assumption is the full list rate.
One more pricing detail: reporting close to launch indicated both models are included across every subscription tier, including the Free plan’s monthly credits — the free tier is enough to evaluate quality, not to run a real workload. The subscription plans range from Starter ($6/month, 30,000 credits) through Creator ($22), Pro ($99), Scale ($299), and Business ($990), so the entry point for serious use is the familiar monthly ladder.
The honest caveats
The discount timing is awkward. If you’re reading this after October 12, assume the discount is gone. If you’re reading it today, don’t rush a purchase decision because of a countdown — evaluate the voice quality first, then decide.
Audio tags bill like spoken words. Every tag character is billed the same as spoken-word characters, so tag-heavy scripts with lots of [whispers] and [laughs] directions cost more than their spoken duration suggests. Not a dealbreaker, but worth knowing when you budget.
Cloning from 10 seconds is a claim, not a proven result. The short-sample instant clone sounds impressive in a launch post. Real-world fidelity — especially carrying tone and personality, not just timbre — is exactly what independent testing will determine in the coming weeks.
Vendor benchmarks aren’t reviews. The #1 arena placement and 75% preference figure come from ElevenLabs. They’re directionally encouraging, but your use case — your language, your voice, your script length — is the only test that counts.
The bottom line
Eleven v4 is a genuinely meaningful release: the narration flagship got longer context, steadier speaker identity, and much wider language coverage, while v4 Turbo gives voice-agent builders a low-latency option that starts speaking in a fraction of a second. The in-script emotion tags are the feature most likely to change how people actually direct AI voices. The 72% launch discount reportedly ends today, so check the live pricing page — but buy for the quality, not the countdown. Evaluate the voices with your own scripts before anything else.
Try ElevenLabs: if this article helped you, you can try the platform here — https://try.elevenlabs.io/haua0qrok9xl — (affiliate link: if you later buy a paid plan through it, I may earn a commission at no extra cost to you).