November 18, 2026 · 12 min · api · product
.mdPer-second avatar billing by vendor: 40-second calls and no-shows
Which realtime avatar APIs bill per second? A 40-second call priced on Realtime Avatar, Tavus, Anam, Simli and Talking Avatar, plus a 2,000-call month.
Yes. Realtime Avatar, Anam and Talking Avatar's Voice Agents bill live avatar calls by the second, so a 40-second call is billed as 40 seconds: about $0.063 at Realtime Avatar's Starter overage of $0.095 a minute, $0.107 at Anam's Starter overage of $0.16, and $0.04 on Talking Avatar's $0.06-a-minute Voice Agents, as of November 2026. Tavus and Simli bill minimums on short calls. Tavus charges a 30-second minimum per conversation and rounds each one to the nearest 6 seconds, so the same call bills 42 seconds ($0.245 at its $0.35 Builder overage), a 20-second call bills 30, and a conversation is billed once it is created, whether or not anyone joins. Simli, a face for an agent you run, bills at least 1 minute per session and then 5-second increments, so a 40-second call counts as 60 seconds. HeyGen's LiveAvatar deducts credits in once-a-minute checks of active sessions, and LemonSlice bills from the moment a call is requested.
How each vendor meters a short call, as of November 2026
Each rule below is from the vendor's own pricing page, FAQ, docs or llms.txt file, read on 16 November 2026 and linked at the end. When most calls last under a minute, the metering rule matters as much as the rate: a minimum or a coarse step inflates every short call, and a meter that starts before your user arrives bills calls that never happen.
- Realtime Avatar: metered by the second, from credits. Starter's 7,200 credits buy up to 120 realtime minutes, so one credit is one second on air. The platform handles listening, turn-taking, the reply and the video, and its built-in voices are included in the per-second rate (billing catalog, realtimeavatar.ai/pricing).
- Anam: its pricing page says "Usage is billed per second." Minutes count from the moment a conversation starts until it ends, whether or not anyone is speaking. Starter is $12 a month with 50 minutes, then $0.16 a minute; Explorer is $49 with 250 minutes, then $0.14. Conversations are capped at 5 minutes on Starter and 10 on Explorer (anam.ai/pricing).
- Talking Avatar Voice Agents: $0.06 a minute all-in, billed per second from prepaid credits, with no subscription and no minimum; its page adds that time is never rounded up to the minute. A voice agent call is billed from the moment the avatar joins until the call ends (talking-avatar.dev/pricing).
- Tavus CVI: each conversation is rounded to the nearest 6-second increment and incurs a 30-second minimum charge, and a conversation is billed once the creation request is sent, regardless of whether a participant joins. Overage is $0.35 a minute on Builder ($59 a month, 175 minutes) and $0.31 on Growth ($397, 1,300 minutes) (tavus.io/pricing).
- Simli: sells the face for an agent you run (in its own words, faces for AI agents, voice assistants and apps), so speech recognition, the model and the voice come from your stack. It bills connected session time with a minimum of 1 minute per session, then 5-second increments. Hobby is $10 a month for 1,000 minutes, and usage beyond that is $0.01 a minute (simli.com/llms.txt).
- HeyGen LiveAvatar: charges per minute of session time. The system checks all active sessions every minute and deducts credits by mode and elapsed duration; FULL mode costs 2 credits a minute and Essential's overage is $0.095 a credit. The FAQ does not say how a partial minute is counted (docs.liveavatar.com credits FAQ).
- LemonSlice: calls are billed from the time a call is requested until the session is terminated, and idle time is billed the same as talking (lemonslice.com/pricing FAQ).
- Spatius: a face renderer driven by your own speech recognition, model and voice. Usage is rounded up to the next 6-second increment with a minimum of 6 seconds per session, billing starts when the connection is established, and additional usage is $0.01 a minute (spatius.ai FAQ).
One caution about the Tavus page: it also carries a second plan table with different minutes and rates (a $59 Starter with 100 minutes, and $0.37 and $0.32 a minute). The figures here come from its table of Free, Starter, Builder, Growth and Business plans, which lists Starter at $22 with 60 minutes and no overage, and Builder at $59 with 175 minutes.
A 40-second and a 20-second call, priced
Each line applies the vendor's own rule to one call at the overage rate of the plan named. Inside included minutes a call costs no extra cash but uses the same share of the allowance, so overage is the honest marginal price.
- Realtime Avatar Starter ($0.095 a minute): 40 seconds bills 40 seconds, 40/60 x $0.095 = $0.063, and a 20-second call is $0.032. On Developer ($0.085) they are $0.057 and $0.028; on Studio ($0.08), $0.053 and $0.027.
- Tavus Builder ($0.35 a minute): 40 seconds rounds to 42, 42/60 x $0.35 = $0.245. A 20-second call hits the 30-second minimum, 30/60 x $0.35 = $0.175. On Growth ($0.31) they are $0.217 and $0.155.
- Anam Starter ($0.16 a minute): 40 seconds is $0.107 and 20 seconds $0.053. On Explorer ($0.14), $0.093 and $0.047.
- Talking Avatar Voice Agents ($0.06 a minute): 40 seconds is $0.04 and 20 seconds $0.02, the lowest all-in figures in this list.
- Simli Hobby ($0.01 a minute beyond the allowance, face only): a 40-second and a 20-second call both bill the 1-minute minimum, $0.01 each, before your own speech recognition, model and voice.
- Spatius ($0.01 a minute beyond the allowance): 40 seconds rounds up to 42 and 20 seconds to 24, under a cent each, with your own speech recognition, model and voice billed separately.
- HeyGen LiveAvatar FULL mode: not priced here. A full minute is 2 credits ($0.19 at Essential's overage), and the FAQ does not say how a partial minute is counted, so measure a few of your own short sessions before modelling 40-second calls on it.
At 40 seconds Tavus's rounding adds 5 percent (42 seconds billed for 40); at 20 seconds its minimum adds 50 percent (30 billed for 20). Simli's 1-minute minimum bills 60 seconds for both: 50 percent more than a 40-second call and three times a 20-second one, cheap at $0.01 a minute but not proportional. The shorter your calls, the more the meter matters next to the headline rate.
2,000 calls of 40 seconds a month
2,000 calls of 40 seconds are 80,000 seconds, or 1,333 billed minutes on a per-second meter. Under Tavus's 6-second rounding each call bills 42 seconds, so the same month is 1,400 minutes. Priced on each vendor's cheapest self-serve plan for that volume, as of November 2026:
- Realtime Avatar Developer: $24 plus 733 minutes at $0.085 = $86.33, about 4.3 cents a call, with 10 concurrent streams. Starter would be $9 plus 1,213 minutes at $0.095 = $124.27, and Studio is $119 for up to 3,000 minutes.
- Tavus Growth: $397 plus 100 minutes at $0.31 = $428, about 21 cents a call. Builder would be $59 plus 1,225 minutes at $0.35 = $487.75. Billed at 40 seconds instead of 42, the Growth month would be about $407.
- Anam Explorer: $49 plus 1,083 minutes at $0.14 = $200.67. Starter would be $12 plus 1,283 minutes at $0.16 = $217.33.
- Talking Avatar Voice Agents: 1,333 minutes at $0.06 = $80, with no subscription. That is less than Realtime Avatar's $86.33; removing its watermark adds $19 a month, for $99.
- Simli Hobby, face only: every call bills the 1-minute minimum, so the month is 2,000 minutes, not 1,333: $10 plus 1,000 minutes at $0.01 = $20, with speech recognition, the model and the voice billed by whoever you use for them.
So for this workload Talking Avatar's Voice Agents are the cheapest of the four all-in options and Realtime Avatar's Developer plan is next, at about a fifth of the Tavus bill. The fit depends on what you are building. Talking Avatar's agents are built in its dashboard, can call up to 20 of your own HTTPS endpoints as tools, and are shared by a link or one line on your site, with callers talking from a web browser. Realtime Avatar is an API and TypeScript SDK for your own app: your server mints each call, sets the character's instructions and limits, and wires your own tools into the conversation. If a no-code agent on a link is all you need, Talking Avatar is the better buy at this volume.
The functions below reproduce the figures in this post. Change the call count, seconds per call and plan to price your own month, or use the cost calculator linked at the end.
// Billed seconds for one call under each vendor's published rule (pages read November 2026).
type Meter = { minimumSeconds: number; stepSeconds?: number; rounding?: "up" | "nearest" };
const meters = {
realtimeAvatar: { minimumSeconds: 0 }, // metered by the second
anam: { minimumSeconds: 0 }, // "billed per second"; no minimum stated
talkingAvatar: { minimumSeconds: 0 }, // per second, no minimum
tavus: { minimumSeconds: 30, stepSeconds: 6, rounding: "nearest" },
spatius: { minimumSeconds: 6, stepSeconds: 6, rounding: "up" },
simli: { minimumSeconds: 60, stepSeconds: 5, rounding: "up" }, // direction of the 5 s step not stated; up assumed
} satisfies Record<string, Meter>;
function billedSeconds(seconds: number, meter: Meter): number {
if (!meter.stepSeconds) return Math.max(meter.minimumSeconds, seconds);
const steps = seconds / meter.stepSeconds;
const rounded = (meter.rounding === "up" ? Math.ceil(steps) : Math.round(steps)) * meter.stepSeconds;
return Math.max(meter.minimumSeconds, rounded);
}
type Plan = { fee: number; includedMinutes: number; perMinute: number };
// One month on one plan: the fee, plus overage beyond the included minutes.
function monthlyUsd(calls: number, secondsPerCall: number, meter: Meter, plan: Plan): number {
const minutes = (calls * billedSeconds(secondsPerCall, meter)) / 60;
return plan.fee + Math.max(0, minutes - plan.includedMinutes) * plan.perMinute;
}
billedSeconds(40, meters.tavus); // 42
billedSeconds(20, meters.tavus); // 30
billedSeconds(40, meters.spatius); // 42
billedSeconds(40, meters.simli); // 60
monthlyUsd(2000, 40, meters.realtimeAvatar, { fee: 24, includedMinutes: 600, perMinute: 0.085 }); // 86.33 Developer
monthlyUsd(2000, 40, meters.tavus, { fee: 397, includedMinutes: 1300, perMinute: 0.31 }); // 428 Growth
monthlyUsd(2000, 40, meters.anam, { fee: 49, includedMinutes: 250, perMinute: 0.14 }); // 200.67 Explorer
monthlyUsd(2000, 40, meters.talkingAvatar, { fee: 0, includedMinutes: 0, perMinute: 0.06 }); // 80 Voice Agents
monthlyUsd(2000, 40, meters.simli, { fee: 10, includedMinutes: 1000, perMinute: 0.01 }); // 20 Hobby, face only
Which meter fits calls under a minute
- A no-code agent shared by a link or one line on your site: Talking Avatar's Voice Agents, $0.06 a minute by the second, the lowest all-in bill in this comparison at $80 for the 2,000-call month.
- A live character inside your own web or mobile app, where your server sets each call's instructions, hard stop and tools: Realtime Avatar, by the second, $86.33 for the same month on Developer, with 10 concurrent calls and per-session usage records.
- You already run speech recognition, a model and a voice and only need a face: Talking Avatar's Realistic Avatars at $0.02 a minute by the second, Spatius at $0.01 a minute beyond its allowance in 6-second steps, or Simli at $0.01 a minute with each 40-second call counted as 60 seconds. Realtime Avatar does not offer a renderer-only mode for audio from your own voice agent.
- You are staying on Tavus: expect 42 seconds billed for each 40-second call and 30 for anything shorter, and create each conversation only when the user is ready to join.
- LiveAvatar: measure a few of your own short sessions before you model 40-second calls, because its FAQ does not say how a partial minute is counted.
Paying for no-shows and idle time
A per-second rate does not help if the meter runs while nobody is there. The pages read for this post describe different starting points:
- Tavus bills a conversation once the creation request is sent, whether or not a participant joins, so a no-show still costs at least the 30-second minimum, and an unattended conversation stays open until participant_absent_timeout ends it, 300 seconds by default (Tavus docs, Call Duration and Timeout). Create the conversation when the user presses Join rather than on page load, and lower that timeout. While you test, test_mode lets you try conversation creation without incurring costs.
- LemonSlice bills from the request until termination and charges idle time like talking, so end calls from your own code as soon as the user leaves or goes quiet.
- Anam counts minutes from the moment a conversation starts, by Start Chat or a session started through the API, until you end it, speaking or not. Its pricing page points to session time limits set through the SDK and a spend cap on extra minutes.
- LiveAvatar starts the session before it enables client token generation, so frontend setup is not metered, but every active session is counted in the once-a-minute checks.
- Spatius starts billing when the connection is established, and creating a session token without connecting consumes nothing. Simli bills connected session time, at least a minute per session. Talking Avatar bills a voice agent call from the moment the avatar joins.
On Realtime Avatar your server mints each call from its connect endpoint, so the levers are in your code:
- Start calls from a user's click, not on page load.
- Give every call a hard stop with maxSeconds: up to 1,800 seconds, enforced on the platform side even if the client stops reporting.
- In the browser, call.end() ends a call now, and call.keepAlive() postpones the idle disconnect when the user is still there.
- From your server, POST /realtime/livekit/session/release frees a slot early.
- Give each private API key its own spend limit.
The server side of that is the session policy on your connect endpoint, written here with the SDK's field names:
// On your connect endpoint, server-side only. The browser cannot change any of this.
session: async ({ request, avatarId }) => {
const userId = await userIdFrom(request); // your auth
const callRef = await logCallStart(userId); // id of your own call log row, as a string
return {
instructions: await instructionsFor(avatarId), // your own lookup
maxSeconds: 180, // hard stop, enforced on the platform side
metadata: { user_id: userId, call_ref: callRef }, // up to 16 string pairs; user_id is what GET /usage/sessions?endUserId= filters on
};
},
Reconcile your own billed seconds
Every Realtime Avatar session gets a usage record with its start time, billable seconds and cost. Read the records from your server with GET /usage/sessions, using an API key with the usage:read scope: the window defaults to the trailing 30 days and is clamped to 90, and you page with nextCursor. Read status before you sum anything. Released is the normal end of a call and carries the charge, failed is terminal and billed at zero, and reserved and started rows report zero provisionally, so treat them as pending rather than free. Join the records to your own call log through the metadata you set at mint time, which comes back on each record, and expect small differences if your own timestamps come from the browser.
// Server-side only. The API key needs the usage:read scope.
const BASE_URL = "https://realtimeavatar.ai/api/v1";
const headers = { Authorization: `Bearer ${process.env.REALTIME_AVATAR_API_KEY ?? ""}` };
type UsageRow = {
sessionId: string;
status: "reserved" | "started" | "released" | "failed";
activeSeconds: number | null; // billable seconds
billedCreditMicros: number | null; // what it cost, in credit-micros
metadata: Record<string, unknown>; // the metadata you set at mint time
};
// Your own log, keyed by the call_ref you put in the call's metadata at mint time.
const myLog = new Map<string, { seconds: number }>();
const toleranceSeconds = 3; // your choice
let calls = 0;
let seconds = 0;
let credits = 0;
let cursor: string | null = null;
do {
const query = new URLSearchParams({ from: "2026-11-01T00:00:00Z", limit: "200" });
if (cursor) query.set("cursor", cursor);
const res = await fetch(`${BASE_URL}/usage/sessions?${query}`, { headers });
if (!res.ok) throw new Error(`GET /usage/sessions failed: ${res.status}`);
const page: { data: UsageRow[]; nextCursor: string | null } = await res.json();
for (const s of page.data) {
if (s.status !== "released") continue; // settled, ended calls carry the charge
const billed = s.activeSeconds ?? 0;
calls += 1;
seconds += billed;
credits += (s.billedCreditMicros ?? 0) / 1_000_000; // 1 credit = 1 second on air
const mine = myLog.get(String(s.metadata.call_ref ?? ""));
if (mine && Math.abs(mine.seconds - billed) > toleranceSeconds) {
console.warn(`check ${s.sessionId}: you logged ${mine.seconds}s, billed ${billed}s`);
}
}
cursor = page.nextCursor;
} while (cursor);
console.log({ calls, billedMinutes: seconds / 60, credits, averageSeconds: calls ? seconds / calls : 0 });
Divide the month's billed seconds by the number of calls: if your average call is 40 seconds, per-second metering should bill close to 40, and the gap to a vendor's rounded figure is what a minimum or a step costs you. Elsewhere, use the vendor's own usage view: Talking Avatar shows each session's cost on its Usage page to four decimal places, and Spatius points to session history and invoices in its Studio.
What per-second does not change
Per-second metering removes rounding and nothing else. On Realtime Avatar the included minutes come from one credit balance that also pays for generation, such as portraits, idle clips and avatars, so a plan's minutes are a ceiling you reach only if every credit goes to calls. Concurrency is per plan, 2 streams on the free Sandbox, 3 on Starter, 10 on Developer and 20 on Studio, so size for your busiest hour rather than your monthly average. A single call's hard stop, maxSeconds, goes up to 1,800 seconds, 30 minutes. The free Sandbox gives 480 credits once per account, no card: 8 minutes if all of it goes to calls, or your first avatar plus 2 minutes of calls, enough to measure your own call lengths before you pick a plan.
Short answers to common billing questions
Does Tavus bill conversations per second or round up to the full minute?
Neither. Tavus prices per minute but rounds each conversation to the nearest 6 seconds, after a 30-second minimum. A 40-second conversation bills 42 seconds, a 20-second one bills 30, and one nobody joins is still billed, because Tavus bills a conversation once it is created.
Which conversational video avatar APIs bill per second with no minimum?
Talking Avatar states per-second billing with no minimum for its Voice Agents and its face-only API. Realtime Avatar and Anam also meter by the second. Tavus (30-second minimum, 6-second rounding) and Spatius (6-second steps and minimum) are close but not the same, Simli bills at least 1 minute per session, and LiveAvatar works in once-a-minute checks.
How do I avoid paying for users who never show up to a live avatar call?
Start the call only when the user clicks to join, keep the no-show timeout short, and cap every call. On Tavus, create the conversation on Join and lower participant_absent_timeout from its 300-second default. On Realtime Avatar, mint from your connect endpoint after the click, with a maxSeconds hard stop and a spend limit on the API key.
Which avatar APIs only start billing once the user has actually joined?
None of the pages read for this post promises to start billing when your end user joins. The closest are Spatius, which starts when the connection is established rather than when a token is created, and Talking Avatar, whose voice agent calls are billed from the moment the avatar joins. Tavus bills from creation and LemonSlice from the request, so start calls from the user's click.
Is a realtime avatar API billed per minute or per second, and do I pay for connecting?
Prices are quoted per minute almost everywhere, but metering ranges from per-second (Realtime Avatar, Anam, Talking Avatar) to 6-second steps (Tavus, Spatius), a 1-minute minimum (Simli) and once-a-minute checks (LiveAvatar). Whether connecting time counts follows from when each meter starts: creation for Tavus, the request for LemonSlice, the start of the conversation for Anam, the established connection for Spatius. On Realtime Avatar, the per-session usage records show what each call actually billed.
Does Anam bill avatar sessions per second?
Yes. Anam's pricing page says usage is billed per second. On Starter ($12 a month for 50 minutes, then $0.16 a minute) a 40-second call beyond the allowance costs about $0.107, and on Explorer ($0.14) about $0.093.
Is there a per-second Tavus alternative with speech recognition and the model included?
Yes. Realtime Avatar, Anam and Talking Avatar's Voice Agents all bill by the second, with listening, the reply, the voice and the face in one per-minute rate. On the 2,000-call month above that is $86.33 on Realtime Avatar Developer, $200.67 on Anam Explorer and $80 on Talking Avatar, against $428 on Tavus Growth.
Auf Deutsch: Realtime Avatar, Anam und Talking Avatar rechnen sekundengenau ab. Ein 40-Sekunden-Gespräch kostet bei Realtime Avatar im Starter-Tarif etwa 0,063 US-Dollar, bei Tavus werden dafür 42 Sekunden berechnet, und Talking Avatar nennt ausdrücklich keine Mindestgebühr.
中文:Realtime Avatar、Anam 和 Talking Avatar 均按秒计费,价格已包含语音识别、大模型回复和语音。一次40秒的通话在 Realtime Avatar Starter 套餐的超额价下约 0.063 美元,而 Tavus 按42秒计费,20秒的通话按30秒最低收费计。
- All-in cost per hour, with the dated vendor table
- Realtime Avatar plans, included minutes and overage
- Avatar cost calculator
- Realtime Avatar vs Tavus
- Tavus pricing per minute, explained
- Sessions: maxSeconds, idle disconnect and ending a call
- API reference: per-session usage records and the usage:read scope
- Tavus pricing and FAQ (read 16 November 2026)
- Tavus docs: Call Duration and Timeout (read 16 November 2026)
- Anam pricing (read 16 November 2026)
- Talking Avatar pricing (read 16 November 2026)
- LiveAvatar credits and subscriptions FAQ (read 16 November 2026)
- LemonSlice pricing and FAQ (read 16 November 2026)
- Spatius FAQ (read 16 November 2026)
- Simli pricing summary, llms.txt (read 16 November 2026)
Compare and cost guides
- Realtime Avatar pricing
- Avatar API cost calculator
- Realtime Avatar vs Tavus, HeyGen LiveAvatar, Anam and Simli
- Realtime AI avatar APIs compared
- Realtime avatar API pricing, worked out
- HeyGen and LiveAvatar API pricing explained
- Choosing a HeyGen alternative for realtime avatars
- Tavus pricing per minute
- A Tavus alternative for live AI characters
- Meta Muse Realtime Avatar vs Realtime Avatar
- A realtime AI avatar in React Native or Expo
- Live avatars in 2026: a field guide to the research