ElevenLabs launches Eleven v4 and v4 Turbo text-to-speech models

ElevenLabs has launched Eleven v4, its most expressive text-to-speech model, alongside Eleven v4 Turbo, a low-latency variant designed for real-time applications. Ranked #1 by Artificial Analysis, Eleven v4 was preferred by ~75% of listeners in blind head-to-head tests against competitors like Cartesia Sonic 3.6 and Google Gemini TTS. The new architecture focuses on interpreting tone, pacing, and context to generate speech that feels natural rather than mechanical.

Performance and Latency

Eleven v4 Turbo combines high expressiveness with speed, achieving a median time to first speech of ~150ms. This is significantly faster than competing models, which ranged from 262ms to 814ms in testing. The model is optimized to work seamlessly with ElevenLabs' ElevenAgents platform, allowing developers to deploy responsive, emotionally intelligent agents in industries such as healthcare and gaming.

Expressiveness and Control

The models support fine-grained control through natural language descriptions and inline tags like [laughs] or [said angrily in French accent]. Eleven v4 follows these direction prompts more accurately than previous versions, enabling precise narration and character performance. Improvements in multi-speaker dynamics allow for more natural dialogue where speakers respond to context rather than delivering isolated lines.

Multilingual and Cloning Improvements

Both models support over 90 languages, with enhanced accent adherence that prevents voices from drifting back to their source accent during generation. Voice cloning fidelity has improved, with Instant Voice Clones now achievable from just 10 seconds of audio. The update also introduces support for Professional Voice Clones (PVC) for the highest-fidelity use cases and ensures consistent speaker identity across long-form content like audiobooks and ads.