ElevenLabs v4: Expressive Voice in 90+ Languages, the New Voice-Agent Standard | inicIA
Direct answer
On September 28, 2026, ElevenLabs launched Eleven v4 and Eleven v4 Turbo: its fourth-generation text-to-speech. What's new: 90+ languages (up from 70), stackable expression tags inline in the script — you write [laughs] or [said angrily in French accent] and the voice performs it — instant voice cloning from just 10 seconds of audio, and v4 Turbo at ~100ms median latency for real-time voice agents. At launch, Artificial Analysis ranked it #1 on its Speech Arena (Elo 1,319, ahead of Cartesia Sonic and Google Gemini Flash TTS) with the highest pronunciation-robustness score ever recorded. List pricing: $80 per million characters (v4) and $40 (Turbo); the launch promo ($22/$11) ended October 12 — evaluate at list price, not the discount.
What changes, in facts
- Direction inside the script: emotion, pacing, pause, and sound-effect tags are written inline and can be stacked in sequence; the model keeps script context to adjust expression over long passages.
- Identity that holds: the narrator or character voice stays consistent across long text and regenerated lines — the weak point of earlier generations.
- 90+ languages with big jumps: the largest quality gains are in Japanese, Brazilian Portuguese, Mandarin, and Cantonese, per the company.
- Honest cross-language cloning: if the target language differs from the reference voice's language, v4 speaks fluently in the target language instead of carrying over the original accent. If your brand voice must keep its accent, test before migrating.
- Two models, two jobs: v4 for quality (audiobooks, ads, characters) and v4 Turbo for real time (voice agents, support, calls).
The new standard to demand from your voice-agent provider
- Directable expression, not flat voice: if your provider cannot direct emotion and pacing inside the script, it is a generation behind.
- Latency measured in milliseconds: ~100ms median inference is the reference for live conversation. Ask for the number, not the adjective.
- Consistency on long calls: the voice cannot "change person" mid-call. Demand a test with your real scripts.
- Per-character pricing at your volumes: at $80 per million characters list, a high-call-volume business must model the monthly cost before signing — and compare alternatives.
Related links
- How Much Does an AI Voice Agent Cost in 2026? Real Prices
- ElevenLabs Is Worth $22 Billion: AI Voice Is the Hottest Enterprise Category
- Meta Muse Now Works for Your Business: Instagram, Ads, and WhatsApp on Assisted Autopilot
Frequently asked questions
What did ElevenLabs launch on September 28, 2026?
Eleven v4 and Eleven v4 Turbo: its fourth-generation text-to-speech, with 90+ languages, stackable expression tags in the script, instant cloning from 10 seconds of audio, and a ~100ms Turbo variant for real time.
What are expression tags?
Directions written inside the script (like [laughs] or tone cues) that steer how each line is delivered: emotion, pacing, pauses, and sound effects, stackable in sequence.
What does it cost?
List price: $80 per million characters (v4) and $40 (Turbo). The launch promotion ($22 and $11) ended October 12, 2026.
How does it compare to competitors?
At launch, Artificial Analysis ranked it #1 on its Speech Arena (Elo 1,319) vs. Cartesia Sonic 3.6 (1,276) and Google Gemini Flash TTS (1,267), with the record pronunciation-robustness score — a third-party measurement, not the company's.
Take the next step
If your business takes calls — sales, appointments, support — the voice your AI speaks with is already your brand. Demand the v4 standard: expression, latency, and consistency. Talk to inicIA on WhatsApp: https://wa.me/17866120930
Chat on WhatsAppBy inicIA
