Nuance Labs, a Seattle AI lab backed by Nvidia, launched Emma on 9 November, an AI persona people can video chat with. Emma runs on Nuplex 1.0, which the company calls “our full-duplex audiovisual foundation model”: it listens and speaks at the same time. In the company’s own research post, Nuplex’s reply timing overlapped a human speaker’s on 89% of 1,785 questions. The next closest system scored 35%. We have not verified the test.
By our arithmetic, 89% of 1,785 is about 1,589 questions, against about 625 for the next closest system, or about 2.5 times as many. The questions come from Nuance’s own recordings of real conversations. Nuance is also opening a credit-metered Nuplex Beta API, so developers can put a talking face into their own apps.
Business Pill · Guardrails
A short Business Engineer explainer on guardrails: rules enforced around an AI model, not inside it. It teaches the general idea only and says nothing about Nuance Labs or Emma.
The key insight: Nuance Labs says Nuplex’s reply timing overlapped a human speaker’s on 89% of 1,785 questions, against 35% for the next closest system. By our arithmetic, that is about 1,589 questions against about 625, or about 2.5 times as many, in a test the company ran on its own recordings.
The research post says Emma is “an AI persona you’ll actually want to meet face-to-face”, made after 18 months of research. Its face and voice are licensed from an actress, the FAQ says, and no account is needed to talk to it. TestingCatalog, which flagged the launch on X, described Emma as free.
Alongside Emma, Nuance released the Nuplex Beta API. Its documentation calls the API Nuance Live: “Build your own real-time app around a Nuance AI persona, with Nuance Live”. The docs list three personas, sam, sonia and mira, and say “The API is in preview: routes and events may change before 1.0”.
A conversation spends credits for its running time, the docs say, and “Your plan sets how much time a credit covers and how many credits you get each day”. The four documentation pages we read give no prices. The persona “speaks English only”, per the limits page.
Share of 1,785 questions where each system’s reply timing overlapped the original human speaker’s, in Nuance Labs’ own test on its own recordings. Source: Nuance Labs research post, 9 November 2026. Company-reported, not independently verified. Question counts are our arithmetic.
What Was Posted on X
The first AI persona youโll actually want to meet face-to-face.
Most AI personas stare blankly while you talk. Emma reacts while youโre still speaking.
After 18 months of research at @nuance_ai, this is Emma, built on Nuplex, our full-duplex audiovisual foundation model.
Fangchang Ma, Nuance Labs’ co-founder and CEO, announcing Emma on X on 9 November. The post calls Nuplex ‘our full-duplex audiovisual foundation model’ and says Emma came after 18 months of research. The piece quotes Nuance’s research post for the test figures.
Nuance Labs has launched Emma, a free AI character that users can video chat with, built on its new Nuplex model.
Emma can listen and talk at the same time, nod, smile, and change her appearance while someone is speaking, like a real person on a video call.
In @nuance_ai tests based on 1,785 recorded questions, Nuplex matched a human’s response time in 89% of cases.
— ๐จ AI News | TestingCatalog (@testingcatalog) Nov 09 20:24
TestingCatalog’s post on X, 9 November, describing Emma as a free AI character for video chat and relaying the 89% figure from Nuance’s tests on 1,785 recorded questions. The figures are Nuance’s own and have not been independently verified.
How Nuance Measured Human Timing
The research post says reply latency “is the number everyone measures, and on its own it doesn’t settle much”. It says open models such as PersonaPlex and Moshi “are tuned to answer fast and max out the existing benchmarks”, while GPT-Live and Gemini Live take longer. “Nuplex sits between them”, the post says.
So Nuance asked “when does a person reply?” It compared when each system started its reply with when the person in the original conversation answered. The post notes that “GPT-Live and Gemini Live were reached through their public APIs, so their timing includes network time”. The sentence giving 35% does not name the system.
The company also hedges its own numbers: “we believe benchmarks in the AV full-duplex space are not accurate to the user experience yet”. Co-founder and CEO Fangchang Ma wrote on X: “We found that pushing our model higher on existing benchmarks made it feel worse to talk to”.
The Limits Nuance Put on Emma
The research post lists rules for Emma and the API. “Emma never claims to be human”, and video served through the API carries a visible watermark. “The Nuplex Beta API can’t clone real people”. The post rules out sexually explicit content and “no AI girlfriends or boyfriends”.
Healthcare and anything involving under-18s, including education, are not in the beta. Emma and the API are “for adults (18+) only”. The FAQ says they are available in “The United States except Colorado, Connecticut, Illinois and Texas; the EU; the UK; and Canada except Quebec”. It gives no reason for those exclusions.
During a call, the post says, automated checks end it if a user expresses intent toward serious violence, raises sexual content or shows weapons on camera. If the persona itself asks for a passcode, card details, money or remote access to a device, “the call ends”.
Where the Rules on AI Companions Stand
The launch lands on the day a draft of the EU’s Digital Fairness Act surfaced. Bloomberg reported that the draft would target “AI systems that simulate personal relationships to encourage users to continue interacting with them”. Reuters reported fines of up to 6% of annual global turnover.
Nuance lists the EU among Emma’s markets. The Commission text is a draft, not law.
The Business Case for an AI Face
Nuance raised a $50 million Series A in September, led by Lightspeed Venture Partners, with NVIDIA and Define Ventures as new investors, its press release says. It named sales and customer service, coaching, professional training and education as use cases. Education is one of the areas the beta now leaves out.
As we read it, Nuance is arguing about the yardstick as much as the score. Its research post says reply latency on its own “doesn’t settle much”, and its co-founder wrote that pushing the model higher on existing benchmarks made it feel worse to talk to. The 89% figure is the company’s answer to that problem, measured by the company.
Nuance Labs, Meet Emma, the First Nuplex Persona, 9 November 2026
“Faster or slower is a product decision, not a score, so we asked a different question: when does a person reply?”
Three Implications
FOR DEVELOPERS The Nuplex Beta API is in preview, English-only and metered in credits, and the docs say routes and events may change before 1.0.
FOR REGULATED USES Healthcare and anything involving under-18s, including education, are outside the beta, and four US states are not on the availability list.
FOR THE EU Bloomberg reported that the draft Digital Fairness Act would target AI systems that simulate personal relationships to keep users interacting; Nuance lists the EU among Emma’s markets.
The framework describes two curves: “The first is what your underlying technology can do. The second is what your users can actually do with it.”
As we read it, Nuance’s post places Emma on the second curve: it says benchmark gains made the model feel worse to talk to, and it measures timing against real conversations instead. The framework does not discuss Nuance Labs.
What Is Not Established
The 89% and 35% figures come from Nuance’s own test on its own recordings, and no outside party has published a replication in the material we read. The research post does not name the 35% system in that sentence. The API prices are not in the documentation pages we read. The 1,589, 625 and 2.5 times figures are our arithmetic.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Nuance Labs launched Emma, a video-chat AI persona, and a Nuplex Beta API on 9 November. In its own test, Nuplex matched human reply timing on 89% of 1,785 questions, about 1,589, against 35%, about 625, for the next closest system, by our arithmetic. Emma is English-only, adults-only, and not offered in four US states.
A note on sourcing. The performance figures come from Nuance Labs’ research post of 9 November, based on the company’s own test and recordings. We have not verified them. The question counts and the 2.5 times ratio are our arithmetic.
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.
