AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Elevenlabs releases two new speech-to-speech models designed for real-time AI conversations

Collected Oct 20, 2026

Elevenlabs has released a pair of speech-to-speech models under the name Eleven Realtime, aimed at real-time AI conversations. The company positions the larger variant, v1 Max, as combining high accuracy with low latency, and says no model currently on the Artificial Analysis leaderboard is both faster and more accurate.

The concrete numbers behind that claim: v1 Max scores 98.8 percent on the Big Bench Audio benchmark and responds in 0.57 seconds. On the Artificial Analysis Speech to Speech leaderboard it sits near the top-left corner, the region where high accuracy and low latency meet. Speech-to-speech models skip the usual pipeline of transcribing audio to text, running a text model, and synthesizing speech back, so latency depends heavily on the model itself; a sub-second figure is the kind of number that makes turn-taking in voice agents feel conversational rather than delayed.

The smaller model, v1 Mini, is built for natural conversations at lower cost. Elevenlabs highlights that it can tell when someone has finished talking versus merely interjecting with something like "mm-hmm" — an end-of-turn judgment that matters in practice, since cutting a speaker off or waiting too long both break the illusion of a real conversation. V1 Max targets more complex tasks such as customer support.

Both models are available through the ElevenAgents platform and through an OpenAI-compatible Realtime API, which lowers the switching cost for teams already building against that interface shape. Pricing is per minute plus token costs, with v1 Max costing more than v1 Mini.

Why it matters: product teams building voice agents now have another hosted option that trades accuracy against cost in two tiers, with the cheaper one tuned for conversational timing. Developers already on OpenAI-compatible realtime APIs, in my reading, face a relatively small integration effort to evaluate a swap.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Elevenlabs' new Eleven Realtime speech-to-speech models combine speed and accuracy. The v1 Max variant scores 98.8 percent on the Big Bench Audio benchmark with a 0.57-second response time. Elevenlabs says no model on the Artificial Analysis leaderboard is both faster and more accurate. The article Elevenlabs releases two new speech-to-speech models designed for real-time AI conversations appeared first on The Decoder .