Thursday, 22 October 2026
Anawords
Research desk
Tools & Apps

ElevenLabs launches Eleven Realtime speech-to-speech models that start replying in about half a second

Max and Mini models are available in ElevenAgents and an OpenAI-compatible Realtime API. ElevenLabs reports 98.8% on Big Bench Audio for Max; pricing is per minute plus model usage.

By Mayank Sahu, Founder & Editor
Sources checked by AI · approved for publication by the author
Published
1 min read · 2 sources
Anawords illustration (not a photo): ElevenLabs Eleven Realtime. Replies in about half a second. Company claim.Illustration · not a photo
Built by Anawords from the facts in this story. Not an official company image. Graphic: Anawords

In short

ElevenLabs launched Eleven Realtime on 20 October 2026, two speech-to-speech models for voice agents. Eleven Realtime v1 Max starts speaking in about 0.57 seconds and scored 98.8% on the Big Bench Audio test, according to ElevenLabs. Both models work through an OpenAI-compatible Realtime API.

ElevenLabs launched Eleven Realtime v1 Max and Mini speech-to-speech models on 20 October 2026.1,2
Both are available in ElevenAgents and an OpenAI-compatible Realtime API.1,2
ElevenLabs reports Max at 0.57 seconds and 98.8% on Big Bench Audio.1,2

What launched

ElevenLabs released two speech-to-speech models, which listen and reply in speech directly rather than converting to text in between:12

  • Eleven Realtime v1 Max: for harder tasks such as customer support, reasoning and multi-step tool calls.12
  • Eleven Realtime v1 Mini: a lower-cost model for natural conversation; it can tell when a speaker has finished or is just saying “mm-hmm”.12

Both are available now in the ElevenAgents platform and through a Realtime API that follows OpenAI’s Realtime format, so developers using that format can switch by changing the endpoint and model name.12

Speed and accuracy

The Decoder reports Max’s response time at 0.57 seconds; artrealmai, citing ElevenLabs’ launch post, gives Mini at 0.62 seconds.12 ElevenLabs reports Max scored 98.8% on Big Bench Audio, a spoken reasoning test, and says no model on the Artificial Analysis speech-to-speech leaderboard is both faster and more accurate.12 These are company-reported results.

Artrealmai also lists 90+ languages, more than 17,000 voices, and tool-use success of 87.1% for Max and 60.3% for Mini.2

Price and limits

Usage is billed per minute plus model costs; artrealmai lists $0.08 per minute for both models plus language-model usage.12 It is a hosted service with no open weights, and the Realtime API supports WebSockets only at launch.2

What it means

Voice agents usually trade speed for accuracy: fast replies feel natural but can get facts wrong, while careful answers create awkward pauses. ElevenLabs says Max narrows that gap. For businesses building phone support or booking agents, OpenAI-format compatibility lowers switching costs, so testing several providers on your own calls is now easier. In India, where many support calls mix languages, testing on real recordings before rollout will matter more than benchmark scores.

What to watch

  • Independent tests on accented and mixed-language speech.
  • Whether WebRTC support arrives for browser and mobile apps.

Related: Mistral’s Voxtral 3 speech models and xAI’s Grok Voice 3.

Keep reading on this topic

Tools & AppsChatGPT now lets you lock chats behind a PIN or Face ID, but they are not encryptedLocked Chats hide sensitive conversations from anyone using your device. They do not change data retention or training settings, and forgetting the PIN means losing the chats.Models & LaunchesMistral launches Voxtral 3 speech models and a voice agent SDK, with Hindi and six other Indian languagesTranscription starts at $0.0035 a minute and covers 50+ languages; speech generation costs $13 per million characters. A LiveKit-based SDK builds voice agents that can run in a company’s own cloud.Models & LaunchesxAI releases Grok Voice 3.0 for voice agents at the same $0.08-a-minute price as beforexAI says the speech-to-speech model answers in under a second and tops an Artificial Analysis ranking. Its own figures show it behind OpenAI’s GPT-Live-1 on one customer-support test.Tools & AppsGitHub makes AI Scan and Agentic Autofix generally available, using AI to find and fix security flaws in codeAI Scan checks pull requests in languages GitHub’s CodeQL does not cover, such as PHP, Bash and Terraform. Agentic Autofix drafts a fix, reruns the scan and opens a pull request for review. More in Tools & Apps →

How this story was checked

  1. ResearchedWith AI assistance · 2 sources
  2. AI source checkEach sentence matched to its source
  3. Editor approvalMayank Sahu · 22 Oct, 19:54
  4. Published22 Oct, 19:54
Read the research behind this story3 sources read · 2 used · what we could not verify · published 22 Oct, 19:44 →

Verify this story

2 reported1 company claim2 unconfirmed

How this was made: researched, drafted and source-checked by AI (Claude, made by Anthropic), then published under the editor’s standing approval. A person did not read it line by line before it went live. AI policy

Fact ledger

How labels work
Reported · 2
ElevenLabs launched Eleven Realtime v1 Max and Mini speech-to-speech models on 20 October 2026.Source 1,2
Both are available in ElevenAgents and an OpenAI-compatible Realtime API.Source 1,2
Company claim · 1
ElevenLabs reports Max at 0.57 seconds and 98.8% on Big Bench Audio.Source 1,2
Unconfirmed · 2
Pricing is $0.08 per minute plus model usage.Source 2
Support for 90+ languages.Source 2

Questions readers ask

What is Eleven Realtime?

A pair of ElevenLabs speech-to-speech models, Max and Mini, for building voice agents that reply quickly.

Is Eleven Realtime compatible with OpenAI’s Realtime API?

Yes. ElevenLabs says it follows OpenAI’s Realtime format, so existing code can switch endpoint and model name.

Sources

  1. 1The Decoder: ElevenLabs releases two new speech-to-speech models for real-time AI conversations
    Checked 22 Oct 2026 · Archived copy (22 Oct 2026)    OUTLET
  2. 2ArtRealm AI: ElevenLabs Eleven Realtime, speech-to-speech voice agents
    Checked 22 Oct 2026 · Archived copy (22 Oct 2026)    OUTLET

Updates and corrections

v122 Oct 2026, 19:54First published. Research note published first. Sources checked by AI; approved for publication by the editor.

Written and approved by

Mayank Sahu, Founder & Editor

Mayank Sahu is an SEO expert who loves using AI and learning how it works. He started Anawords because he felt the market was missing the best AI articles: clear, sourced and honest about what is confirmed and what is only claimed. He sets the rules every Anawords story must pass, runs its AI-assisted research desk and gives the approval its stories are published under. Each story says how it was checked.

ProfileLinkedInFact-check policyAI policy
See a mistake, or named in this story? We reply within 24 hours. Report an error