ElevenLabs Releases Eleven Flash Ultra-Low Latency Model
ElevenLabs launched Eleven Flash (v2.5), a real-time text-to-speech model offering execution latencies as low as 75 milliseconds. The model costs $0.015 per 1,000 characters and supports generation across 32 languages.
Verified State Diff
Impact & Verification Analysis
Developers and enterprise teams building real-time voice agents, customer service automation, and interactive gaming experiences.
Reduces end-to-end voice response latency to near-human conversational levels while cutting API compute costs in half, making real-time voice AI deployment economically viable at scale.
Full Fact Overview
Eleven Flash is engineered specifically for latency-sensitive applications such as conversational agents, live customer service bots, and interactive game NPCs. Operating with a 75ms generation delay, Flash delivers a 3x speed improvement over Turbo v2.5 while cutting token API costs by 50%. The model supports streaming output via WebSockets and HTTP API endpoints using the model_id `eleven_flash_v2_5`, while maintaining high audio fidelity and voice-cloning accuracy.