Back to the blog

November 18, 2026 · 11 min · api · product

.md

AI avatar cost per user, student or conversation: worked examples

50 hours of live AI avatar calls: $119 a month on Realtime Avatar, billed by the second. Worked costs per user, student and conversation, vs Tavus and Anam.

Fifty hours of live AI avatar calls a month is 3,000 minutes. On Realtime Avatar, which meters calls by the second, that is exactly the Studio plan's 3,000 included minutes: $119 a month as of November 2026, about 4 cents a minute or $2.38 an hour, against $228 on the Developer plan and $282.60 on Starter. Divide the $119 by your unit: $1.19 per user for 100 users who each talk 30 minutes a month, $2.38 per student for 50 students at an hour a month each, or about 16 cents per conversation for 750 four-minute conversations. Because billing is per second, a call that lasts 4 minutes 10 seconds uses 250 credits, one per second, not five rounded-up minutes.

At other volumes the same arithmetic gives $9.58 per user per month for 50 users who each talk to the avatar for 5 minutes a day (7,500 minutes, $479 on Studio), $6.53 per student for a 300-student academy at 20 minutes a week, about 26 cents per 4-minute support conversation at 2,000 a month, and about 12 cents per 2-minute product-page conversation. For comparison, the same 50 hours cost $924 on Tavus Growth, $475 on HeyGen LiveAvatar Business in FULL mode, $419 on Anam Growth and $180 on Talking Avatar's Voice Agents, each priced from the vendor's own pricing page as read on 16 November 2026. A face-only renderer such as Simli streams 3,000 minutes for about $30 on its $10 Hobby plan, but then speech recognition, the language model and the voice are yours to run and pay for.

The three-line formula, and where each plan wins

Every cost question here reduces to three lines: monthly minutes from your own usage, a bill on a plan, and that bill divided by your unit, whether a user, a student or a conversation. Realtime Avatar sells minutes as credits: Starter's 7,200 credits buy up to 120 realtime minutes, so one credit is one second on air.

monthly minutes = users x minutes per session x sessions per month
monthly bill    = plan fee + max(0, monthly minutes - included minutes) x overage rate
cost per unit   = monthly bill / units (users, students or conversations)

The self-serve plans, from the billing catalog the pricing page reads, as of November 2026:

  • Starter: $9 a month, 120 minutes included, then $0.095 a minute ($5.70 an hour), 3 concurrent streams, 5 avatars included.
  • Developer: $24 a month, 600 minutes included, then $0.085 a minute ($5.10 an hour), 10 concurrent streams, 25 avatars included.
  • Studio: $119 a month, 3,000 minutes included, then $0.08 a minute ($4.80 an hour), 20 concurrent streams, 100 avatars included.
  • Sandbox: free, 480 credits once per account, no card: 8 minutes if every credit goes to calls, or a first avatar plus 2 minutes of calls.
  • Scale: for high volume, with reserved warm capacity and credits sized to your volume; ask the team for its terms.

Two break-even points decide the plan. Starter is cheaper than Developer only below about 278 minutes a month: its 120 included minutes plus the $15 price gap divided by $0.095. Developer is cheaper than Studio only below about 1,718 minutes: 600 included plus the $95 gap divided by $0.085. Above 1,718 minutes Studio is the lowest-cost of the three, and because its overage is also the lowest of the three, it stays ahead of Starter and Developer at every higher volume.

  • Up to about 278 minutes a month (about 4.6 hours): Starter, $9 plus $0.095 a minute beyond its 120 minutes.
  • About 278 to 1,718 minutes (up to about 29 hours): Developer, $24 plus $0.085 a minute beyond its 600 minutes.
  • Above about 1,718 minutes: Studio, $119 plus $0.08 a minute beyond its 3,000 minutes; 50 hours lands exactly on its allowance.
  • A busy hour with more than 20 calls at once: ask about Scale, whatever the monthly minutes.

Minutes are not the only limit. Each plan also caps simultaneous calls (3, 10 and 20 streams), so pick the cheapest plan whose streams cover your busiest moment, not your average. Average concurrency is small: 18,000 minutes in a 30-day month is 600 minutes a day, under one call at a time across a 12-hour day. Peaks are what bite, such as a class that starts on the hour or a flash sale on a product page. Above 20 simultaneous calls, look at Scale.

Worked examples per user, per student and per conversation

Each example uses a 30-day month, 52 weeks over 12 months for weekly habits, prices as of November 2026, and the plan the formula says is cheapest. They assume every included credit goes to calls.

  • 50 hours a month, any app: 3,000 minutes, exactly Studio's allowance, so $119 ($2.38 an hour). Developer would be $24 plus 2,400 minutes at $0.085, or $228, and Starter $9 plus 2,880 minutes at $0.095, or $282.60.
  • A SaaS built in Lovable where 50 users each talk to the avatar for 5 minutes a day: 50 x 5 x 30 = 7,500 minutes. Studio is $119 plus 4,500 minutes at $0.08, so $479 a month, $9.58 per user and about 6.4 cents a minute blended. Developer would be $610.50.
  • An AI conversation tutor, one student doing three 15-minute sessions a week: 45 minutes a week is about 195 minutes a month. At Studio's $0.08 overage that student adds $15.60 a month. A tutoring app with 100 such students uses 19,500 minutes, $1,439 on Studio or $14.39 per student. One student on their own fits Starter at $16.13.
  • An English academy in Mexico with 300 students who each speak for 20 minutes a week: 300 x 20 x 52 / 12 = 26,000 minutes. Studio is $119 plus 23,000 minutes at $0.08, so $1,959 a month or $6.53 per student. If whole classes start at the same time, check that peak against Studio's 20 concurrent streams, and at this volume ask about Scale too.
  • A support desk with 2,000 conversations a month at about 4 minutes each: 8,000 minutes. Studio is $119 plus 5,000 minutes at $0.08, so $519, or about $0.26 per conversation. With per-second metering, a 90-second password reset costs 90 seconds. The same 8,000 minutes on the vendors listed below: Tavus $2,015, LiveAvatar about $1,375, Anam $999, Talking Avatar's Voice Agents $480.
  • A product expert on product pages, where 3% of 100,000 monthly visitors talk for 2 minutes: 3,000 conversations and 6,000 minutes. Studio is $119 plus 3,000 minutes at $0.08, so $359, about $0.12 per conversation or $3.59 per 1,000 visitors.
  • An online store taking 200 live consultations a day at 3 minutes each, for example a Korean shopping mall with an avatar on its product detail pages: 200 x 3 x 30 = 18,000 minutes. Studio is $119 plus 15,000 minutes at $0.08, so $1,319 a month, or about $0.22 per consultation.

The same volumes on Tavus, HeyGen LiveAvatar, Anam, Talking Avatar and Simli

These figures were read from each vendor's own page on 16 November 2026 (links at the end) and priced with the same formula on each vendor's cheapest self-serve plan for the volume. Tavus, LiveAvatar's FULL mode and Talking Avatar's Voice Agents state that speech recognition, the language model and the voice are included, as they are in Realtime Avatar's rate; Anam counts the whole session, speaking or not. Recheck each page before you commit.

  • Realtime Avatar (billing catalog, November 2026): 3,000 minutes cost $119, 8,000 minutes $519 and 18,000 minutes $1,319, all on Studio, billed by the second, with 20 concurrent streams.
  • Tavus (tavus.io/pricing, read 16 November 2026): Growth is $397 a month with 1,300 minutes, then $0.31 a minute; Business is $975 with 4,000 minutes, then $0.26. That puts 3,000 minutes at $924 on Growth, and 8,000 minutes at $2,015 and 18,000 at $4,615 on Business (Growth would be $5,574 at 18,000). Growth allows 10 concurrent sessions and Business 15. Each conversation has a 30-second minimum and is rounded to the nearest 6 seconds.
  • HeyGen LiveAvatar, FULL mode (liveavatar.com and its credits documentation, read 16 November 2026): Business is $475 a month for 6,000 credits, FULL mode uses 2 credits a minute, and overage is $0.09 a credit, which is $0.18 a minute. 3,000 minutes fit inside Business for $475; 8,000 minutes come to about $1,375 and 18,000 to about $3,175. Its credits documentation says it charges per minute of session time. LiveAvatar lists unlimited concurrency on its paid plans, with sessions of up to 60 minutes on Business.
  • Anam (anam.ai/pricing, read 16 November 2026): Growth is $299 a month with 2,000 minutes, then $0.12; Professional is $999 with 8,000 minutes, then $0.11. 3,000 minutes cost $419 on Growth, and 8,000 minutes cost $999 and 18,000 cost $2,099 on Professional. Growth allows 5 simultaneous sessions and Professional 10.
  • Talking Avatar Voice Agents (talking-avatar.dev/pricing, read 16 November 2026): $0.06 a minute all-in, billed per second from prepaid credits with no subscription. 3,000 minutes cost $180, 8,000 cost $480 and 18,000 cost $1,080. Removing its watermark is $19 a month more, and it quotes custom pricing above 50,000 minutes a month.
  • Simli, face only (simli.com/pricing, read 16 November 2026): Hobby is $10 a month with 1,000 minutes at $0.01 a minute, Pro $49 with 5,500 at $0.0095 and Scale $249 with 27,500 at $0.009. That puts 3,000 minutes at about $30 on Hobby, and 18,000 minutes at about $168 on Pro. Simli describes itself as a speech-to-video API that adds faces to realtime AI agents, so the agent behind the face, meaning speech recognition, the LLM and the voice, is yours to run and pay for.

Where another product fits better: Talking Avatar's Voice Agents cost less than Studio above about 6,050 minutes a month (about 7,000 if you also pay its $19 a month to remove the watermark), so they beat us on the 8,000-minute support desk and the 18,000-minute store. Without the watermark removal they also undercut our cheapest plan below about 400 minutes a month and between about 1,080 and 1,980 minutes; with it, Realtime Avatar costs less at every volume below about 7,000 minutes. Talking Avatar's agents are built in its dashboard and shared by link, widget or iframe, which suits a team that wants a hosted agent rather than code in its own app. Face-only renderers such as Simli are far cheaper per minute because the conversation is billed elsewhere. Realtime Avatar offers no renderer-only mode for audio from your own voice agent, so if you already run one on LiveKit Agents or Pipecat, a face-only renderer is the honest fit. Far beyond 20 simultaneous calls, LiveAvatar's unlimited concurrency matters more than its rate.

Pricing your own app on top

Margin per user is your price minus the minutes you include times your marginal rate. On Studio the marginal rate is $0.08 a minute ($4.80 an hour) once the 3,000 included minutes are used; before that, the next minute is already paid for.

  • A minute allowance per plan: if your $15-a-month tier includes 60 avatar minutes, the worst case at $0.08 is $4.80 of usage, leaving $10.20 before your other costs. Most users will not use all of it, but price for the ones who do.
  • Flat fee or per-minute resale: a flat fee is simpler when usage per user is predictable, as with a tutor booked three times a week. Per-minute resale protects you when a few users talk far more than the rest, as with an open support widget. A small allowance followed by a per-minute price covers both.
  • Cap each call: your server sets maxSeconds when it starts a call, so a 4-minute cap bounds one support conversation at 32 cents on Studio's overage rate. Before you share an app publicly, require a signed-in user in the connect route, because every call spends your balance.
  • Re-bill from the records: each session's usage record carries its start time, billable seconds and cost, which is what you need to charge your own customers by the second or to audit a month.
  • New characters: beyond a plan's included avatars (5 on Starter, 25 on Developer, 100 on Studio), each extra avatar costs $3.

The function below picks the cheapest plan for a month of usage and divides by your unit. Swap in your own users, minutes and peak concurrency; it returns undefined when your peak needs more than Studio's 20 streams.

// Realtime Avatar self-serve plans as of November 2026 (realtimeavatar.ai/pricing).
const plans = [
  { name: "Starter", fee: 9, included: 120, overage: 0.095, streams: 3 },
  { name: "Developer", fee: 24, included: 600, overage: 0.085, streams: 10 },
  { name: "Studio", fee: 119, included: 3000, overage: 0.08, streams: 20 },
];

// Cheapest plan for a month of usage. undefined = a peak above 20 streams (ask about Scale).
function monthlyBill(minutes: number, peakStreams = 1) {
  return plans
    .filter((plan) => plan.streams >= peakStreams)
    .map((plan) => {
      const usd = plan.fee + Math.max(0, minutes - plan.included) * plan.overage;
      return { plan: plan.name, usd: Math.round(usd * 100) / 100 };
    })
    .sort((a, b) => a.usd - b.usd)[0];
}

// 50 users x 5 minutes a day x 30 days, at most 5 calls at once.
const users = 50;
const bill = monthlyBill(users * 5 * 30, 5);
if (bill) console.log(`${bill.plan}: $${bill.usd}, $${(bill.usd / users).toFixed(2)} per user`);
// Studio: $479, $9.58 per user

// Billed by the second: a 4 min 10 s call is 250 seconds.
const callUsd = (seconds: number, perMinute = 0.08) => (seconds / 60) * perMinute;
console.log(callUsd(250).toFixed(2)); // "0.33" at Studio's overage rate

The per-call cap lives in your server's session policy, next to the character's instructions:

// Inside your connect route's session callback: the server sets the cap, the browser cannot.
session: async ({ avatarId }) => ({
  instructions: await promptFor(avatarId), // your own lookup
  maxSeconds: 4 * 60, // a 4-minute cap: at most $0.32 at Studio's $0.08 overage
}),

What the per-minute rate covers, and what it does not

The Realtime Avatar rate is the whole call, not a face price waiting for an LLM bill. The platform handles listening, turn-taking, interruption, the reply and the video, and its built-in voices are included in the per-second rate. Two options are billed by their own providers. Your own text-to-speech plugs in through a signed voice webhook (realtime-avatar SDK 0.28 or later). Separately, an OpenAI-compatible model endpoint you host can answer the turns; external endpoints are enabled per workspace on request.

Credits are one balance. The credits that pay for call seconds also pay for generation, such as portraits, idle clips and avatars, so a plan's included minutes are a ceiling you reach only if every credit goes to calls; leave headroom if you generate often. Your own hosting, your backend and the APIs your tools call are outside the rate.

Cost questions builders ask

How much would 5,000 minutes a month cost on HeyGen LiveAvatar?

In FULL mode, 5,000 minutes are 10,000 credits. The Business plan's $475 covers 6,000 of them and the other 4,000 at $0.09 add $360, so about $835 a month as of 16 November 2026; Essential would be about $945. The same 5,000 minutes on Realtime Avatar are $279 on Studio: $119 plus 2,000 minutes at $0.08.

D-ID Agents or a per-second API for an AI tutor at 1,000 student minutes a month?

On Realtime Avatar, 1,000 minutes sits between the two break-even points, so Developer is the cheapest plan: $24 plus 400 minutes at $0.085, or $58 a month. This post does not price D-ID Agents; take D-ID's current rates from its own pricing page and run 1,000 minutes through the same formula. For reference at that volume, Talking Avatar's Voice Agents would be $60, and Anam's Explorer plan $154, with conversations capped at 10 minutes, shorter than a 15-minute lesson.

What would a Portuguese-language online store pay for 3,000 three-minute conversations?

That is 9,000 minutes: $119 plus 6,000 minutes at $0.08, so $599 a month on Studio, or about $0.20 per conversation. Every figure in this post is minutes and money only; test your callers' language in a free Sandbox call before you commit.

Check your own numbers

Put your users, session length and sessions per month into the avatar cost calculator, compare the dated vendor table on the pricing page, and try a real call on the free Sandbox: 480 credits once per account, 8 minutes if all of it goes to calls. The AI tutor and support avatar use-case pages show those integrations.

Compare and cost guides

Meet the live avatars. Hold the first conversation.

Enter the studio