ElevenLabs introduced Eleven Realtime on October 20, 2026. It describes it as “a new class of Speech to Speech models” for voice agents, where speech recognition, turn-taking, the language model and the voice are tuned together instead of chained.
What to know
- Two models: v1 Mini is the lower-cost option, aimed at natural, expressive conversation such as assistants, tutors and interactive characters. ElevenLabs lists 620ms time to first audio for it. v1 Max is aimed at complex tasks such as customer support, lead qualification and multi-step workflows with tool calls, with 570ms time to first audio listed.
- Benchmarks (ElevenLabs’ own figures): Mini: 95.8% on Full-Duplex-Bench and an Elo of 988 on Speech Agent Arena. Max: 98.8% on Big Bench Audio and an 87.1% task success rate. Wait for independent tests.
- Languages: realtime conversation in over 90 languages, with a range of accents.
- Where to use it: pick v1 Mini or v1 Max from an agent’s model dropdown in ElevenAgents, which can deploy to phone, web widget, WhatsApp and contact centre platforms. There is also a Realtime API, compatible with OpenAI’s Realtime API, so you change the endpoint and model name.
- Limits at launch: the API connects over WebSocket only (no WebRTC), and ElevenAgents features such as channels, knowledge base, guardrails, testing and monitoring are not available through the Realtime API.
- Pricing: ElevenLabs lists per-minute pricing plus token costs but gives no rates on the page, so check its pricing page.
Why it matters
Fewer moving parts can mean simpler voice agents, but you still need a good prompt and real test calls. Start with how to write voice agent prompts, compare with Mistral’s Voxtral 3 and Google’s Gemini 3.8 Live, and see the ElevenLabs page and the model tracker.
Source: ElevenLabs. We summarise in our own words; check the source for the full details.