Updates·October 21, 2026, 06:46

ElevenLabs launches two voice models that can hold real-time conversations

AI-generated from the sources listed, with automated source checks.

ElevenLabs has released Eleven Realtime in two versions, designed to make AI voice agents sound more natural. The larger model, v1 Max, scores highest on reasoning but also costs more than the smaller one.

On October 20, 2026, ElevenLabs launched Eleven Realtime, a new type of model for voice agents. A voice agent is an AI you can talk to on the phone or online, and it answers in spoken language. The models are so-called speech-to-speech models, meaning they work directly from speech to speech.

The models come in two versions: v1 Mini and v1 Max.

Who is each for?

v1 Mini is aimed at AI assistants, tutors, interactive characters and most everyday use cases. v1 Max is meant for customer service, lead qualification (that is, assessing whether a potential customer is of interest), multistep workflows and complex reasoning.

The difference shows in the figures, which ElevenLabs reports itself. On Big Bench Audio, where the models are given 1,000 spoken reasoning questions, v1 Max answered correctly 98.8% of the time and v1 Mini 75.9%. When the models have to complete tasks using tools, v1 Max scores 87.1% and v1 Mini 60.3%.

On conversational dynamics, measured with Full-Duplex-Bench, it is the other way around: v1 Mini scores 95.8% and v1 Max 94.1%. In Speech Agent Arena, which measures human preference, v1 Max gets 998 and v1 Mini 988.

Fast responses

Time to first audio averages 0.57 seconds for v1 Max and 0.62 seconds for v1 Mini. According to ElevenLabs, no model on the Artificial Analysis leaderboard is both faster and more accurate than v1 Max. The technology news site The Decoder writes that v1 Max sits close to the upper left corner of the leaderboard, meaning high accuracy and low latency.

The figures come from ElevenLabs' own launch and were, according to ElevenLabs, measured by Artificial Analysis. The Decoder largely relies on the same information.

Languages, voices and pricing

The models support more than 90 languages and can use more than 17,000 voices from ElevenLabs' voice library. They can be used through the ElevenAgents platform and an OpenAI-compatible Realtime API with WebSocket. The channels are phone, web, WhatsApp and CCaaS.

Pricing is calculated per minute plus tokens, and v1 Max costs more than v1 Mini. The actual prices are not given in the sources.

For everyday users, this means customer service and other voice services could become faster and more natural to talk to if companies adopt the models.