ElevenLabs introduced Eleven Realtime on October 20, 2026, a family of speech-to-speech models that packages speech recognition, turn-taking, an LLM, and text-to-speech into a single system. Realtime v1 Max answered 98.8% of Big Bench Audio's spoken reasoning questions and starts speaking in 0.57 seconds, while the lower-cost Mini targets assistants, tutors, and interactive characters.
On October 20, 2026, ElevenLabs introduced Eleven Realtime, a new family of speech-to-speech models that bundles an entire voice agent into a single system. Two variants launched: v1 Mini for natural conversation at low cost, positioned for AI assistants, tutors, and interactive characters, and v1 Max for complex tasks such as customer support, lead qualification, and multi-step workflows where the agent has to follow a process and reason through steps. Both models support more than 90 languages and can use any of the 17,000+ voices in the ElevenLabs library.
Each model combines speech recognition, voice activity detection, turn-taking, an underlying LLM, and text-to-speech into one packaged system, removing the component-by-component assembly that voice agent builders previously had to tune by hand. In independent evaluations by Artificial Analysis across three public benchmarks, v1 Max answered 98.8% of the 1,000 spoken reasoning questions on Big Bench Audio correctly, versus 75.9% for Mini, and started speaking in 0.57 seconds on average against Mini's 0.62 seconds. Mini scored 95.8% on Full-Duplex-Bench — which tests pauses, turn-taking, interruptions, and backchannels — ahead of Max at 94.1%, while Max led the Speech Agent Arena with a preference Elo of 998 to Mini's 988 and completed 87.1% of tool-use tasks versus Mini's 60.3%.
Developers already building on OpenAI's Realtime API can switch with minimal friction: the ElevenLabs Realtime API follows OpenAI's Realtime specifications over WebSocket, so the OpenAI SDK works as-is and migration means changing the endpoint and model name. WebRTC integrations must move to WebSocket first, however. The models are also available through ElevenAgents, the company's agent platform, which adds production deployment channels — phone, web, WhatsApp, and contact-center platforms — plus workflows, knowledge bases, guardrails, testing, and monitoring.
Both models are priced at $0.08 per minute plus LLM usage, a base of $4.80 per connected hour before token costs — the same per-minute rate SpaceXSI set for Grok Voice 3.0 when it launched a week earlier. In its announcement, ElevenLabs framed the goal as removing configuration burden: "You shouldn't have to become a configuration expert to build a voice agent that sounds natural." The launch deepens an already crowded race in real-time voice agents, where OpenAI, SpaceXSI, and Mistral are all competing on latency, reasoning, and platform tooling.
