Skip to content
Neural Daily

ModelsNews

Eleven Realtime: speech-to-speech models for voice agents

ElevenLabs has released two speech-to-speech models for voice agents. v1 Mini targets fast, natural conversation and v1 Max targets complex tasks with tool calls. Both speak the OpenAI Realtime API protocol.

By: Mikola Máté Patrik · · 2 min read

ElevenLabs Eleven Realtime cover with two spheres labeled Max and Mini and the text Instant conversation
Image: ElevenLabs · Press kit · Official product page social image (og:image) · Source

What it is

Eleven Realtime (ElevenLabs) gives you a complete voice agent from one model choice: speech recognition, voice activity detection and turn-taking, an LLM and text to speech, tuned by ElevenLabs as one system. Until now developers picked and wired these pieces separately.

The two models

  • v1 Mini (eleven_realtime_v1_mini): for assistants, tutors and game characters. ElevenLabs reports 620 ms to first audio and 95.8% on Full-Duplex-Bench.
  • v1 Max (eleven_realtime_v1_max): reasons before it responds, for support and multi-step tool calling. 570 ms to first audio, 98.8% on Big Bench Audio and an 87.1% task success rate.
  • Both support 90+ languages, with voices from the ElevenLabs voice library.

For developers

The models are selectable from a dropdown in ElevenAgents, or reachable over a WebSocket API that uses the same client and server events as OpenAI's Realtime API. The docs say an existing app built on OpenAI's Realtime SDK can point at it without switching SDKs.

There are limits: Mini is not available under Zero Retention Mode or HIPAA, Max is not offered in the India and Singapore data residency environments and needs a token limit of at least 1,024. Pricing is $0.08 per minute plus LLM usage, with Max costing more per token.

The figures are vendor benchmarks. If you are assembling your own ASR, LLM and TTS chain, a quick Mini prototype is a cheap way to compare latency.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles