ElevenLabs launches Eleven Realtime speech-to-speech models for voice agents

On October 20, 2026, ElevenLabs launched Eleven Realtime, speech-to-speech models in two versions, v1 Mini and v1 Max, with a Realtime API that ElevenLabs says is compatible with OpenAI's Realtime API.

Archive entry: this covers an event from October 20, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 20, 2026, ElevenLabs launched Eleven Realtime, a class of speech-to-speech models for voice agents, and described it in its announcement on the ElevenLabs blog. The launch models were v1 Mini, for most use cases, and v1 Max, for multi-step processes and heavy tool calling.

What ElevenLabs launched

Both models support 90+ languages and any voice in ElevenLabs' catalog of 17,000+ voices.
Eleven Realtime is built on a cascaded architecture tuned end to end. Every step produces text, which enables ElevenAgents guardrails and PII redaction.
ElevenLabs says its Realtime API is compatible with OpenAI's Realtime API, and switching requires changing the endpoint and the model identifier.
The Realtime API connects over WebSocket only. The example endpoint is wss://api.elevenlabs.io/v1/convai/realtime with the model eleven_realtime_v1_mini.
ElevenLabs reported that v1 Max started speaking in 0.57 seconds on average and v1 Mini in 0.62 seconds.

What it means for people who build with AI

Developers with code written against OpenAI's Realtime API could try ElevenLabs voices and models for voice agents by swapping the endpoint and the model ID. Because every step of the cascaded pipeline produces text, ElevenAgents guardrails and PII redaction could be applied, and the two versions let teams choose between v1 Mini for most use cases and v1 Max for multi-step processes with heavy tool calling.

Availability and cited results

The models were available in ElevenAgents and the Realtime API, and ElevenAgents offered phone, web, WhatsApp and CCaaS channels. The announcement did not state per-minute pricing or plan availability. It also cited Artificial Analysis results: Big Bench Audio at 98.8% for v1 Max and 75.9% for v1 Mini, Full-Duplex-Bench at 95.8% for v1 Mini and 94.1% for v1 Max, and tool use at 87.1% for v1 Max and 60.3% for v1 Mini.

Sources

Introducing Eleven Realtime: a simpler way to build natural voice agents (ElevenLabs, October 20, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.