ElevenLabs introduced Eleven Realtime on 20 October 2026, which it calls “a new class of Speech to Speech models” for voice agents. The company says its v1 Max model starts speaking in 0.57 seconds on average, and that the models are available through a Realtime API that is compatible with OpenAI’s.
“Our fastest yet for voice agents,” ElevenLabs CEO Mati Staniszewski wrote on X, saying Eleven Realtime “listens, thinks and replies in under half a second.”
Business Pill · LATENCY VS THROUGHPUT
A short explainer of latency versus throughput. It teaches the general idea only and says nothing about ElevenLabs or its models.
The key insight: As we read it, ElevenLabs is pitching speed and accuracy together, and it removes a switching step by making its Realtime API compatible with OpenAI’s.
The launch post on X says voice agents “usually make you choose between quick replies or precise answers,” and that Eleven Realtime “gives you both.” ElevenLabs offers two versions, v1 Mini and v1 Max.
ElevenLabs’ product page says v1 Mini is “optimized for natural conversations” and is aimed at AI assistants, educational tutors and interactive characters. It says v1 Max is “optimized for complex tasks” such as customer support and lead qualification, and that it reasons through multi-step requests and calls tools.
The company says speech recognition, turn-taking, the LLM and voice are “tuned end-to-end” as one system, and that conversation works in over 90 languages.
Benchmark scores ElevenLabs reports for Eleven Realtime: 98.8% on Big Bench Audio and an 87.1% task success rate for v1 Max, 95.8% on Full-Duplex-Bench for v1 Mini; time to first audio 570ms (Max) and 620ms (Mini), the average measured on Big Bench Audio. ElevenLabs credits the evaluations to Artificial Analysis. Company figures, not FourWeekMBA tests. Source: ElevenLabs.
What Was Posted on X
Introducing Eleven Realtime, a new class of Speech to Speech models.
Voice agents usually make you choose between quick replies or precise answers.
Eleven Realtime gives you both. It starts talking in half a second and gets spoken reasoning questions right.
ElevenLabs posted the launch on X at 14:25 UTC on 20 October, calling Eleven Realtime a new class of Speech to Speech models. The post says voice agents usually make you choose between quick replies or precise answers, and that Eleven Realtime starts talking in half a second. The benchmark figures in this piece come from the company’s product page.
ElevenLabs CEO Mati Staniszewski posted at 15:18 UTC that Eleven Realtime is the company’s fastest model yet for voice agents and that it listens, thinks and replies in under half a second. These are the company’s claims; we have not tested the models.
The Numbers ElevenLabs Gives
ElevenLabs’ page lists a time to first audio of 570ms for v1 Max and 620ms for v1 Mini. It lists 98.8% on Big Bench Audio and an 87.1% task success rate for v1 Max, and 95.8% on Full-Duplex-Bench and a Speech Agent Arena Elo of 988 for v1 Mini.
The page says Artificial Analysis evaluated both models independently on those benchmarks, and that time to first audio is the average measured on Big Bench Audio. ElevenLabs’ launch thread says that “No model on the Artificial Analysis leaderboard is both faster and more accurate” than v1 Max. We have not seen the leaderboard itself, and these are the company’s figures.
How Developers Get It
ElevenLabs says both models are available in ElevenAgents, where a developer selects Eleven Realtime as the model and can deploy the same agent to phone lines, websites, WhatsApp and contact-center platforms.
The product page says the Realtime API is compatible with OpenAI’s Realtime API, so developers “change the endpoint and model name and keep the rest of your code.” It adds that every client and server event “arrives as it does from OpenAI.”
On pricing, the page says that in ElevenAgents, Realtime models use standard per-minute pricing plus the LLM cost per token, and that through the Realtime API “you pay per minute of connected audio plus tokens.” It points to its pricing page for current rates and does not list them on the product page.
OpenAI has also been wiring decisions into voice: its changelog entry of 15 October says it added a native Decisions integration in GPT-Live, which we covered in our Decisions API piece. This week, Hark said it delivered nearly 20,000 pizzas despite technical problems caused by overwhelming demand during Hark’s free-pizza promo, Axios reported, and Microsoft said its security tools now discover 60 types of local AI agents, as we covered in Microsoft’s count of 60 agent types.
The Structural Read
As we read it, the launch thread frames the trade-off directly: it says voice agents usually make you choose between quick replies or precise answers, and that Eleven Realtime gives you both.
As we read it, the integration pitch is about code reuse: the product page says developers change the endpoint and model name and keep the rest of their code.
As we read it, the figures are the company’s: ElevenLabs credits them to Artificial Analysis evaluations and says time to first audio is the average measured on Big Bench Audio.
Mati Staniszewski, CEO, ElevenLabs, on X, 20 October 2026
“Our fastest yet for voice agents”
Three Implications
TWO MODELS, TWO JOBS ElevenLabs says v1 Mini is optimized for natural conversations and v1 Max for complex tasks such as customer support and lead qualification.
A DROP-IN API The product page says the Realtime API is compatible with OpenAI’s, with the same client and server events.
COMPANY FIGURES ElevenLabs lists 570ms to first audio for v1 Max and 620ms for v1 Mini, credited to Artificial Analysis evaluations.
The framework puts it this way: “The surface, not the model, governs autonomy.”
As we read it, ElevenLabs is selling a surface as well as a model: its page says speech recognition, turn-taking, the LLM and voice are tuned end-to-end, and that agents deploy to phone, web, WhatsApp and contact centers from ElevenAgents. The framework was published in July 2026 and does not discuss ElevenLabs.
What Is Not Established
The benchmark figures are ElevenLabs’ own as published on its page, credited to Artificial Analysis evaluations we have not reviewed. The product page does not state per-minute prices, and we have not tested either model.
Business Engineer Framework
The Agentic Harness War
A Business Engineer analysis of where value pools in the agent stack.
ElevenLabs launched Eleven Realtime on 20 October 2026 in two versions, v1 Mini and v1 Max. It says v1 Max starts speaking in 0.57 seconds on average and scores 98.8% on Big Bench Audio, and that both models run through ElevenAgents and a Realtime API compatible with OpenAI’s. Those figures are the company’s.
A note on sourcing. This piece is built on ElevenLabs’ Eleven Realtime product page, read in full, its launch thread on X and its CEO’s post, all of 20 October 2026. Every benchmark and latency figure is the company’s own, credited by ElevenLabs to Artificial Analysis; we have not tested the models. Nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
