Why we removed LiveAvatar's concurrency limits and took prices down to $0.01 a minute.
Starting today, LiveAvatar offers unlimited concurrency. You can run one avatar or ten thousand at the same time: same API, no slots to reserve, no capacity calls with sales. And with the new pricing, full-body and half-body 1080p avatars stream for as low as $0.01 per minute at scale*.
Here's the story behind it — starting with the constraint most builders only discover in production: how many sessions you can run at once.
The invisible ceiling
Our industry grew up selling scarcity, ourselves included. Avatar slots. Session caps. Reserved capacity sold per seat, per stream, per month. There was a reason: real-time video generation was genuinely expensive, GPUs were scarce, and fencing off capacity per customer was the safe way to guarantee quality.
But rationed presence breaks real user experience, because real traffic is spiky.
A fleet of retail kiosks peaks at lunch. A classroom logs in at the same minute. A launch brings a month of traffic in one afternoon. If your platform caps you at forty concurrent sessions, your product has a ceiling of forty simultaneous conversations, and you hit it at exactly your best moments.
Cloud computing went through this same transition: reserved capacity gave way to elastic, pay-for-what-you-use compute, and nobody went back. Avatars lagged for a physical reason — a real-time 1080p human isn't a stateless function you can cold-start in milliseconds. It's a persistent, GPU-heavy, latency-critical stream, and making that elastic is genuinely hard. That's the problem we've been engineering against, and today it works the way electricity does:
Nobody meters your outlets. We don't meter your concurrency.
Presence is simply there when your users arrive, whether that's one of them or a hundred thousand. And to be precise about what unlimited means: there is no cap on simultaneous sessions on any paid plan*. Start as many as your users need. You pay for minutes streamed, not for the right to run them.
Demand was never the problem
Analysts called this market years ago. Gartner has projected a digital human economy worth roughly $125 billion by 2035, and predicted that half of B2B buyers would interact with a digital human in their buying cycle by 2026 — a shift we're helping make, not just watching. Here's what we're seeing with our AI sales agents:
https://x.com/wayne_liang_/status/2084320195112485276
Where the economics already work, demand isn't theoretical: in China, e-commerce platforms run AI hosts around the clock. JD.com reports its virtual livestreamers have promoted 4,000+ brands, cut hosting costs by roughly 90% versus human-run rooms, and lifted off-peak conversion by 30%. And when entrepreneur Luo Yonghao put his AI twin on a Baidu livestream, it drew 13 million viewers and over $7 million in sales in six hours, beating his own human-hosted record.
https://x.com/CNBC/status/1935634378597498904
Here's the lesson we take from it:
Face-to-face AI isn't really competing with other avatars. It's competing with no avatar.
Voice-only bots, chat windows, static FAQs, or nothing at all. And what wins those users isn't one more point of photorealism. It's scalability with cost-efficiency.
One nuance: livestream commerce is one-to-many: a single avatar broadcasting to millions, one lane of real-time avatars. The lane we're building for is more demanding: one-to-one, every user has their own private personalized conversation. Broadcast needs one session regardless of audience; interactive needs as many sessions as you have users. That's exactly where cost and concurrency bind, and exactly what this launch removes.
So we made scalability the product
The thing we're proudest of this year isn't a model release. It's a curve.
For the past six months, our infrastructure team has chased a single number: concurrent sessions per server at full quality — the thousandth stream held to the same 1080p, the same frame rate, the same latency as the first.
That curve has no single breakthrough behind it. It's six months of unglamorous engineering stacked layer on layer: session scheduling that packs streams onto servers more tightly, batching that lets one GPU render many conversations at once, a generation model made leaner so the same quality takes less compute, and warm capacity pools so a new session starts instantly instead of waiting on a cold GPU. None of it demos well. All of it compounds: the same server that carried two simultaneous streams in March carries thirty-two today.
And a 16× denser server pays out twice. Capacity stopped being scarce, so we stopped rationing it. Minutes stopped costing what they did, so we stopped charging what we did.
Unlimited concurrency and $0.01 minutes aren't two announcements. They're one engineering result, read two ways.
There's precedent for what happens when this curve gets pushed hard. Stanford's AI Index found that inference at GPT-3.5-level quality fell more than 280-fold in price in under two years — and usage exploded. Nobody spent less on AI because tokens got cheap; whole product categories became possible that weren't before. We believe real-time avatar minutes follow the same curve, and we intend to be the ones driving it.
For context on where today's pricing lands: published rates across managed real-time avatar platforms run $0.10–$0.37 per active minute, and self-hosting runs $0.06–$0.12 per minute of compute before ops overhead. At the top tier, a professional-grade avatar on LiveAvatar now streams for less than the raw GPU math of building one yourself.
The economics of good enough
HeyGen built its reputation on avatar realism, it's what we're known for in video, and the research bar we hold ourselves to everywhere. Real-time turned out to be a different discipline. The question our customers actually ask isn't "how much more realistic can it get?" It's "how do I make the math work to put a real-time avatar in my product?"
Customers, including some who left, kept telling us the same thing:
Price is the number-one reason customers have left us — 29%, nearly a third. Realism doesn't appear on the list.
Realism is a threshold, not a ladder. Past the threshold, it's economics.
So we prioritize in the order builders experience it: cheap enough to try on day one, cost-efficient enough to dare to scale.
Good enough to build trust is where the outcomes live. edYOU calls LiveAvatar "the voice and face the student sees, that's where trust and engagement happen." Master English sees learners who practice with an avatar walk into real conversations feeling confident. Zeligate's candidates put it more personally: "I feel heard, I'm not being judged for my accent."
Every one of those outcomes came from an avatar people trusted, delivered at a price and scale where it could actually be deployed.
Run the math at the new rate
The numbers below use the $0.01/min tier rate. That's what the math looks like once you're at scale. And remember, until today, concurrency itself was a second bill.
A school
Give every one of a school's 1,000 students fifteen minutes a day of face-to-face tutoring with an avatar — a daily dose our customers have shown moves grades meaningfully. At $0.01 a minute, that's about $150 a day for the entire school. The same minutes from human tutors: more than $10,000 a day at typical US rates. Same minutes, same students, 67× cheaper. And the old second bill: reserving capacity for 1,000 simultaneous sessions used to run about $10,000 a month, streamed or not. Today that line item is $0.
A hiring platform
Say a hiring platform gives every applicant a real 20-minute first-round interview instead of screening most out by résumé. Interviewing all 100,000 applicants costs $20,000 in avatar time, versus well over $1M in recruiter labor, if a human team even could. A 50× difference. And the interview-day spike of 500 simultaneous sessions, once a $5,000-a-month reservation, is now $0.
A retail chain
Retailers are already piloting avatar assistants and holographic hosts in stores. Skechers alone has run an in-store AI shopping assistant and, this spring, a hologram brand-ambassador experience. The realistic pattern isn't a 24/7 live avatar: looped product clips, with a live session starting only when a shopper walks up. Sixty conversations a day at four minutes each is $2.40 a day per store, versus roughly $135 a day for one associate's shift, about 2% of the cost. And when the lunch rush hits every store at once, there's no capacity to pre-buy, the sessions just start.
Across all three:
A personalized human presence for every user isn't a line item anymore. It's the default.
What we believe
We started LiveAvatar because we think human presence should be a primitive of AI software — something any builder can call, the way they call compute or storage. Everything in this launch follows from that. We'll keep making it more affordable and scalable, keep the quality professional-grade, and keep sharing what we learn as we go.
The curve keeps going.
Presence, unlimited. Come build with us: www.liveavatar.com
- \Pricing note: $0.01/min is the unit price at the large-volume Enterprise tier. [Contact sales if you are interested](https://www.liveavatar.com/contact-sales).*
- \Concurrency note: Planning something enormous — tens of thousands of simultaneous sessions? Tell us and we'll pre-warm capacity for you.*
Appendix
- Stanford HAI: 2025 AI Index Report: GPT-3.5-level inference cost fell 280x ($20 → $0.07 per million tokens) between Nov 2022 and Oct 2024. https://hai.stanford.edu/ai-index/2025-ai-index-report
- Gartner: emerging-technology research (2023): digital human economy projected at ~$125B by 2035; 50% of B2B buyers interacting with a digital human in the buying cycle by 2026. Gartner's research is subscription-only; figures attributed to the firm's Hype Cycle analysis as widely reported.
- China Daily HK: JD.com virtual livestreamers: 4,000+ brands, ~90% cost reduction vs. human-run rooms, +30% off-peak order conversion. https://www.chinadailyhk.com/hk/article/582576
- Skechers retail pilots: Proto hologram brand-ambassador experience (2026): https://retailtechinnovationhub.com/home/2026/4/5/skechers-blends-entertainment-tech-and-retail-with-new-proto-hologram-howie-mandel-experience
- LiveAvatar customer stories: https://www.liveavatar.com/customer-story/edyou, https://www.liveavatar.com/customer-story/zeligate, https://www.liveavatar.com/customer-story/master-english
