ElevenLabs launches v4 speech model with 10,000-character limit and Turbo variant
ElevenLabs released Eleven v4, a speech model that better follows tone, pacing, and context cues while keeping voices consistent across long productions. The update supports over 90 languages, up from about 70 in v3, and handles up to 10,000 characters per request—roughly ten minutes of audio. Users can now direct pronunciation via phonetic spelling or tags, and cloned voices retain native…
Key points
- ElevenLabs’ v4 model improves tone, pacing, and pronunciation consistency with phonetic spelling support and 90+ languages
- v4 Turbo responds in **150 ms**, faster than Cartesia Sonic 3.6 (**262 ms**) and OpenAI’s GPT-4o mini TTS (**814 ms**)
- Temporary pricing cuts: **$22** for v4 (from **$80**) and **$11** for Turbo (from **$40**) until October 12
ElevenLabs also introduced Eleven v4 Turbo, a faster variant optimized for real-time voice agents like customer service or game characters. It starts producing speech in 150 milliseconds, compared to 262 ms for Cartesia Sonic 3.6 and 814 ms for OpenAI’s GPT-4o mini TTS. The company claims v4 ranks ahead of competitors on Artificial Analysis’ Provider Voice Arena leaderboard, scoring 91.7% on pronunciation benchmarks—up from 85.6% in v3. Blind tests showed about three-quarters of listeners preferred v4 over models from Cartesia, Inworld, and Google.
Both models launch with temporary pricing cuts: $22 per million characters for v4 (down from $80) and $11 for Turbo (down from $40). Users on the $22 Creator plan or higher can access v4 in ElevenCreative for two weeks at no extra cost, with usage capped at twice their monthly credits. Enterprise customers can store data in isolated EU, India, or Singapore environments, though some processing may occur outside the chosen region.
Model pages: Eleven v4 → · Gemini 3.8 Flash TTS →
The story so far
2 episodes →- ElevenLabs launches v4 speech model with 10,000-character limit and Turbo variantthis story
ElevenLabs' new v4 speech model makes AI voices more expressive and consistent
The Decoder · 29 September 2026
Loading the full article…
This text was published by The Decoder and written by Jonathan Kemper. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- AI agents automate tasks across tools without human input · 2 src
- McKinsey survey finds only 10% of AI agent experiments scale beyond pilot stage · 1 src
- Meta adds Zapier connector to Muse agent for 9,000+ apps · 1 src
- Meta Muse agent allegedly sold item and shared address without permission · 4 src
- TypeSafe AI launches Jev, a decision-focused AI 193x faster and 444x cheaper than ChatGPT · 5 src
Comments
via GitHub Discussions