ElevenLabs demoed Eleven Realtime v1 Mini and v1 Max on Oct 20, two speech-to-speech models that replace a multi-component voice stack with audio in, audio out and a single model ID. The models target agents built in ElevenLabs ElevenAgents and through its Realtime API, and they ship as product releases, not security patches.
Before Oct 20, building a natural agent meant choosing five parts. Teams picked speech recognition, voice activity detection, turn-taking, an LLM and text to speech, then tuned the handoffs themselves.
That tuning is now bundled. ElevenLabs said you "shouldn't have to become a configuration expert to build a voice agent that sounds natural," a line aimed at teams wrestling with pause length, interruptions and voice and prompt interactions. The company said ElevenAgents already exposed those settings, and its Realtime API is compatible with OpenAI's Realtime API so existing code needs only an endpoint and model name change.
Both models support more than 90 languages and any of the 17,000 plus voices in the ElevenLabs library, according to the company. Eleven Realtime is built on a cascaded architecture tuned end to end, so it can be measured on the same leaderboards as fused and duplex models while keeping text at every step.
That text matters in production. You can see what the agent heard and what it decided to say, which separates a misheard word from a wrong decision, and it lets ElevenAgents apply guardrails and PII redaction to a Realtime agent, the company said.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 64,000 other subscribers.
Artificial Analysis evaluated both models independently, according to ElevenLabs. On 1,000 Big Bench Audio reasoning questions, v1 Max answered 98.8% correctly and v1 Mini 75.9%. V1 Max started speaking in 0.57 seconds on average and v1 Mini in 0.62 seconds, and no model on the Artificial Analysis leaderboard was both faster and more accurate than v1 Max, the company said. On Full-Duplex-Bench for pauses and turn-taking, v1 Mini scored 95.8% and v1 Max 94.1%. In the Speech Agent Arena blind conversations, v1 Max scored 998 and v1 Mini 988, with tool-use task success at 87.1% for v1 Max and 60.3% for v1 Mini.
Availability is live. In ElevenAgents you select a Realtime model from the dropdown and keep phone, web, WhatsApp and CCaaS channels, workflows, knowledge base, guardrails, testing and monitoring. Through the Realtime API you connect over WebSocket with the OpenAI SDK, as shown in the docs example for eleven_realtime_v1_mini with outputModalities audio and voice brian. WebRTC integrations must move to WebSocket first, and teams that need channels or a knowledge base must build on ElevenAgents, with the migration guide and docs published as of Oct 29.
