Back to the blog

October 5, 2026 · 9 min · api · pricing · comparison

.md

Realtime AI avatar APIs compared (October 2026): price per minute, free tiers, SDKs, tools

Tavus, LiveAvatar, Anam, Simli, Beyond Presence, D-ID, bitHuman and Realtime Avatar compared on price per minute, free tier, SDKs and tool calling.

Pick by architecture first, then by price. If you already run a voice agent (LiveKit Agents, Pipecat, OpenAI Realtime, ElevenLabs), buy a renderer: HeyGen LiveAvatar in Avatar Only mode, Anam, Simli, bitHuman or Beyond Presence's speech-to-video API. If you want one API that listens, thinks, speaks and renders, the full-stack options in October 2026 range from about $4.80 an hour (Realtime Avatar, Studio overage) and $6 an hour (bitHuman managed voice chat at the top-up rate) to $15.60 to $21 an hour (Tavus overage), with Anam, LiveAvatar, Beyond Presence and D-ID in between.

Every competitor number below was read from that vendor's own pricing page or documentation on 5 October 2026 and is labelled as of October 2026. Sources are linked at the end. Where a vendor does not publish a number, the table says so instead of borrowing one from a review site. We make Realtime Avatar, so read our rows with that in mind, and check our claims against our own pricing page, which is the source for them.

Price, free tier and pipeline, as of October 2026

The marginal rate is the per-minute price after a plan's included minutes run out, because that is the number that decides your bill once people actually use the product. Where a vendor lists several plans, the rates are in plan order from cheapest plan to most expensive. LiveAvatar's Full Mode uses 2 credits a minute, so $0.095 and $0.09 per credit become $0.19 and $0.18 per minute. Beyond Presence prices in euros.

API (product)            Paid entry (USD/EUR per month)   Marginal live rate, full conversation   Free tier                                Pipeline you get
Tavus CVI                Starter $22 (60 min, no overage)  $0.35 / $0.31 / $0.26 per min           20 min/month, 5 min per call            Full stack, or your LLM; Pipecat/LiveKit modes
HeyGen LiveAvatar        Essential $99 (1,100 credits)     $0.19 / $0.18 per min (Full Mode)       10 credits, 2 min per session           Full Mode, or Avatar Only (your agent)
Anam                     Starter $12 (50 min)              $0.16 / $0.14 / $0.12 / $0.11 per min   30 min/month, 3 min per call            Turnkey, or bring LLM / TTS / audio
Simli                    not published                     not published                           $10 on signup + 50 min monthly top-up   Renderer: you bring STT, LLM, TTS
Beyond Presence          Starter EUR 49 (14K credits)      EUR 0.35 / 0.20 / 0.175 per agent min    2K credits/month (20 agent min)         Managed agents, or speech-to-video
D-ID Agents              Build $18 (up to 32 stream min)   no overage rate published               14-day trial, up to 10 stream min       Full stack, or your LLM / your audio
bitHuman                 Creator $20 (2,000 credits)       $0.10 per min managed voice (top-up)    Featured avatars; API needs Creator     Renderer (cloud or device), or managed voice
Realtime Avatar          Starter $9 (120 min)              $0.095 / $0.085 / $0.08 per min         1,020 credits once (17 min)             Full stack only (STT, LLM, voice, video)

Read the last column before the price column. A renderer turns audio you produce into a talking face; you still pay separately for speech recognition, a language model and a voice. A full stack sells the whole conversation. bitHuman's cloud rendering alone is 4 credits a minute ($0.04 at its $1 = 100 credits top-up rate), which is cheap because it does less; its managed voice chat at 10 credits a minute is the comparable full-stack number. LiveAvatar's Avatar Only mode is 1 credit a minute ($0.095 or $0.09) for the same reason.

SDKs and tool calling

API                 SDKs and client surfaces (from official docs)                                  Tool calling
Tavus CVI           @tavus/cvi-ui (React, web); Daily SDKs for React Native, iOS, Android, Flutter  Yes: tools registry, delivered by app message or HTTPS call
HeyGen LiveAvatar   @heygen/liveavatar-web-sdk; LiveKit, Pipecat, Agora plugins                     Not documented for Full Mode; via your own agent in Avatar Only
Anam                JavaScript SDK, Python SDK, LiveKit and Pipecat; community Flutter and KMP      Yes: client tools and webhook tools
Simli               JavaScript and Python SDKs; LiveKit and Pipecat                                 In your own agent stack
Beyond Presence     Any LiveKit client SDK (web, mobile); LiveKit plugin (Python, JS); iframe       Function calling listed from the Scale plan
D-ID Agents         @d-id/client-sdk (web), embed script; LiveKit SDK for V4 Expressive agents      Yes: server tools and client tools
bitHuman            Swift package, Android SDK, Python, CLI, LiveKit, Pipecat, web embed            In your agent framework
Realtime Avatar     realtime-avatar on npm: browser, React, React Native, server adapters; OpenAPI  Yes: client tools in your page (2.5 s budget), server steering

React Native is where the field thins out. Tavus documents React Native through Daily's React Native SDK joining the same conversation URL. Beyond Presence and D-ID's V4 agents stream over LiveKit, so LiveKit's own client SDKs are an option, and bitHuman ships native Swift and Android SDKs rather than a React Native package. Realtime Avatar ships a first-party realtime-avatar/react-native entry point, though it is lower-level than the web component; the React and React Native docs show the exact code.

Billing details that change the real cost

  • Tavus bills each conversation a 30-second minimum and rounds to the nearest 6 seconds. Free and Starter calls are capped at 5 minutes, Builder at 15, Growth and Business at 60. Recordings cost $0.03 a minute.
  • Anam meters the whole session, talking or idle, by the second; included minutes do not roll over. Per-call limits are 3 minutes on Free, 5 on Starter, 10 on Explorer and 2 hours on Growth and Professional.
  • LiveAvatar deducts credits per minute of session time and bills overage in $100 increments. Essential sessions are capped at 20 minutes, Business at 60. Paid plans advertise unlimited concurrency.
  • bitHuman counts active session time, talking or idle, billed to the second. Credits bought with a plan are cheaper than top-ups: Business is $299 for 50,000 credits, which works out to about $0.06 a minute for managed voice chat if you use the whole allowance. From 12 October 2026 the API and SDKs require the Creator plan or higher.
  • D-ID's Build plan is a personal license with a D-ID watermark; commercial use starts at Launch ($50 a month for up to 90 streaming minutes). Its plan math lands at roughly $0.50 to $0.56 per streaming minute on monthly billing.
  • Realtime Avatar meters seconds on air. One call can run at most 1,800 seconds (30 minutes), enforced server-side, which is shorter than Tavus Growth, LiveAvatar Business or Anam Growth allow. If your product needs hour-long sessions, that matters.

Latency: what each vendor claims

None of these numbers measure the same thing, and none of them are our measurements. They are each vendor's own published claim, quoted so you know what to test. Tavus says its Phoenix rendering model has 134 ms latency and its LLMs reply in about 600 ms. Anam says 180 ms average agent response time. Beyond Presence says 250 ms or less global avatar latency, with managed agents streaming at about 1.0 to 1.2 seconds. LiveAvatar says under 300 ms median time to first frame. Simli says under 300 ms for its speech-to-video stage and notes that STT, LLM and TTS add their own time. D-ID's Agents page says answers arrive in under two seconds.

Realtime Avatar publishes two server-side design targets, not measured guarantees: 300 ms to speech and 500 ms to the first video frame. They are not an SLA, and no build is gated on them. If latency decides your choice, run the same short script against each finalist; our benchmarking post describes a protocol.

Short verdicts by use case

You already have a voice agent

Buy a renderer, not a second pipeline. LiveAvatar Avatar Only has connectors for ElevenLabs, Cartesia, OpenAI Realtime and Gemini Live plus LiveKit, Pipecat and Agora plugins. Anam, Simli, bitHuman and Beyond Presence all have LiveKit plugins, and Tavus documents Pipecat and LiveKit modes. Realtime Avatar is the wrong choice here: it has no LiveKit or Pipecat plugin and cannot render audio from your own voice agent. You can switch off its speech recognition and drive each turn with text from your own LLM, but the voice is still ours.

You want the most human-looking replica and enterprise paperwork

Tavus is the strongest fit. It trains faces from a two-minute video or a photo, adds perception (Raven-1) and turn-taking (Sparrow-2) models, joins Google Meet and Zoom, and lists SOC 2 and HIPAA compliance. You pay for it: $15.60 to $21 an hour at overage.

Long sessions or EU-first procurement

Anam's Growth plan allows 2-hour conversations at $0.12 a minute overage. Beyond Presence is built in Europe and lists GDPR, EU AI Act and SOC 2 Type II on its site, with on-premise deployment on Enterprise.

On-device rendering

bitHuman is the only one here that renders on the device (iPhone, iPad, Mac, Android arm64, Linux, WebGPU browsers), which also makes it the cheapest renderer at 2 credits a minute on your own hardware.

A character in your own web or React Native app, with your tools, at low per-hour cost

This is the case Realtime Avatar is built for. You create a character from one portrait, mint each call from your own server route, render it with one React component (or the React Native primitives), and the character calls functions that run in your page with your credentials. The whole conversation is $4.80 to $5.70 an hour at overage, plans start at $9 a month for 120 minutes, and the free sandbox is 1,020 credits granted once per account, not monthly. The trade-offs: no renderer mode, Fish Audio voices only, a session-selectable LLM from our own set (OpenAI or Gemini models) rather than your own endpoint, and a 30-minute cap per call.

How to decide in an afternoon

  1. Write down whether you own the conversation already. If yes, shortlist renderers only.
  2. Estimate monthly live minutes and peak concurrent calls, then price each finalist at its marginal rate, not its headline plan.
  3. Check the per-call cap against your longest real session.
  4. Run one identical scripted conversation on each free tier and note time to first reply, interruption handling and how tools behave.

Prices in this category change often. Tavus's pricing page was serving two different developer price tables in the same page on 5 October 2026 (the second shows Starter at $59 with $0.37 a minute overage); the table above uses the one listing Builder and Business. Re-check every number, including ours, before you commit.

Meet the live avatars. Hold the first conversation.

Enter the studio