ElevenLabs Launches V4, Turbo Voice Models

// 2h agoMODEL RELEASE

ElevenLabs Launches V4, Turbo Voice Models

ElevenLabs launched new expressive text-to-speech models built on a new architecture, supporting 90+ languages, stronger speaker consistency, inline performance controls, and Professional Voice Clones. V4 Turbo targets real-time agents with roughly 100 ms median inference latency and streaming support.

// ANALYSIS

ElevenLabs is pushing voice AI toward both studio-grade performance and genuinely conversational agents. The compelling differentiator is combining emotional control with production-ready latency, though developers should distinguish model latency from total end-to-end agent response time.

–Eleven v4 is best suited to narration, dubbing, audiobooks, character dialogue, and other produced content where delivery matters.
–V4 Turbo brings comparable expressiveness to live voice agents through bidirectional streaming and approximately 150 ms time to first speech.
–Inline tags for emotion, pacing, reactions, and sound effects give developers more deterministic control than prompt-only direction.
–Improved speaker stability, multi-speaker dialogue, and Professional Voice Clone support address common weaknesses in long-form generation.
–The 90+ language coverage broadens deployment options, particularly for multilingual customer support and localized media.

// TAGS

eleven-v4eleven-v4-turbottsspeechvoice-agentapi

DISCOVERED

2h ago

2026-10-02

PUBLISHED

7h ago

2026-10-02

RELEVANCE

9/ 10

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

RESEARCH32m ago

RADAR Detects, Tames Reasoning Loops

A new paper introduces RADAR, a real-time method for detecting when large reasoning models drift from productive reflection into redundant or persistent generation loops. Attention realignment reduces these failures while largely preserving normal performance.

RESEARCH32m ago

CICM exposes stale binding in LLMs

UC Berkeley researchers introduce CICM, a benchmark showing that LLMs can retain updated facts yet answer with outdated values because attention selects stale mentions. The paper proposes training-free attention redirection that corrects most tested errors.

RESEARCH32m ago

Long-Transduction exposes long-horizon agent failures

NVIDIA researchers introduce Long-Transduction, a controlled diagnostic for testing whether models can repeatedly read, mutate, and output state-dependent results across long contexts. Across open-weight models, performance fell 62.8% as context expanded from 4K to 128K, with additional losses from input-format variation and task complexity.