Synthesia API in 2026: Interactive Avatar Pricing, Credits and Limits

One API key covers three products. The Video API renders finished presenter videos from a script, document or template and hands you an .mp4; on self-serve plans it comes with Pro (up to 360 minutes of video a year) and Enterprise, and Synthesia's help center prices it at 160 credits per minute of API usage, against about 100 for a minute made in the Studio editor. The Dubbing API translates existing videos for $0.40 a minute, or $0.80 with lip sync. Interactive Avatars are the real-time part: a hosted avatar that joins your LiveKit room and lip-syncs the speech your own agent produces. Only that third product competes with Talking Avatar, which is built for real time. If a recorded training or explainer video, or dubbing, is what you need, Synthesia is the tool and this page won't change that. The rest of this comparison is about live, two-way conversations with an avatar.

Interactive Avatar sessions use 10 credits per minute, which Synthesia's API rate card and developer docs list as $0.10 a minute, as of 30 October 2026. There is no separate API bill: "API usage draws from the same credit balance as your subscription." So the real price depends on the plan. Basic is free with 500 credits a month (about 50 minutes). Starter is $19 a month for 1,750 credits, Pro $89 for 6,500, and Pro 2× to 8× $176 to $676 for 12,500 to 48,500. Divide and a minute costs $0.109 on Starter, $0.137 on Pro and about $0.14 on the larger Pro levels, or $0.08–0.09 billed yearly. Credits reset each billing period and don't roll over; top-up prices aren't published, and neither is how partial minutes are counted. Synthesia's own pages also disagree in places: its Interactive Avatars page shows $0.12 a minute pay-as-you-go and a launch offer of 500 free minutes. We use the rate card's $0.10.

At 1,000 minutes a month you need 10,000 credits. The cheapest self-serve plan that covers that is Pro 2× at $176 a month (12,500 credits, so 1,250 minutes), or $117 a month billed yearly. The same minutes cost $20 on Talking Avatar Realistic Avatars and $60 on Talking Avatar Voice Agents, which include the LLM, speech recognition and voice. At 10,000 minutes you need 100,000 credits a month. The largest self-serve level, Pro 8× at $676, holds 48,500 (4,850 minutes), so this volume means an Enterprise quote or top-ups; at the $0.10 rate card it would be $1,000. Realistic Avatars costs $200, and our enterprise and high-volume contracts go as low as $0.005 a minute. When Synthesia's full stack ships at 20 credits a minute, 1,000 minutes would need 20,000 credits, which is Pro 4× at $347 a month.

You attach Synthesia's AvatarSession to a LiveKit voice agent. A hosted worker joins your room as a participant, receives your agent's audio and publishes the lip-synced avatar. The plugin exists for Python (on PyPI since 15 September 2026) and Node.js (on npm since 9 October); without it you call the REST API and manage LiveKit tokens yourself. Synthesia's docs list what isn't there yet: LiveKit is the only transport, there's no embed or iframe snippet, no session list, transcripts, recordings or webhooks, framing is bust only, and mobile isn't a primary target. Concurrency is 1 session on the free plan and 100 on paid plans, with a separate minute cap whose threshold isn't published. Synthesia publishes no latency figure, and its plugin reference says the avatar has typically taken about 45 seconds to join a room, against a 30-second default timeout. One thing it does well: a session can load up to five avatars and swap between them without reconnecting.

Synthesia's library is large for rendered video (125+ avatars on Starter, 240+ on Enterprise), but its docs say stock actor-based avatars, such as Ryan or Ada, can't be used as Interactive Avatars. Only personal avatars (a digital twin of a real person: 3 on Starter, 5 on Pro, unlimited on Enterprise) and synthetic ones, including the Style Avatars you build in Avatar Builder, can be converted for real-time use through the API. On the free Basic plan, which has no personal avatars, that narrows the choice. Talking Avatar Realistic Avatars starts from a growing library of ready-made photorealistic avatars, casual and professional, that work in real time from the first API call. If you want a specific person, a custom avatar costs $199 once, with the presenter's consent, and then bills at the same $0.02 a minute.

Both companies split real-time avatars into two offers. Synthesia's avatar-only tier ($0.10 a minute) expects your own LLM, speech recognition and voice, and its full stack, with Synthesia's voice and LLM, is listed at $0.20 a minute and marked coming soon on 30 October 2026, although its July launch post promised it for 1 October. Talking Avatar sells both today. Talking Avatar Realistic Avatars is the avatar for the agent you already run (OpenAI Realtime, LiveKit Agents, Pipecat or any TTS) at $0.02 a minute, streamed to your page over WebRTC or into your LiveKit room. Talking Avatar Voice Agents are the complete agent at $0.06 a minute: you write a prompt, pick an avatar and a voice, add up to 20 tools that call your URLs, and share it as a link or embed it with one line. Every call gets a transcript and an optional post-call webhook.

On Synthesia's avatar-only tier your agent already makes the audio, so switching changes only the face. Keep your rendered videos in Synthesia if you use them; this replaces the live avatar.

1. Create a Talking Avatar Realistic Avatars account (it comes with $1 of free credits) and an API key.
2. Pick a ready-made avatar, or order a custom one for $199, one time, with the presenter's consent.
3. Remove Synthesia's AvatarSession from your agent and send the same audio to our avatar, in your LiveKit room or over WebRTC to your page.
4. Keep your LLM, speech recognition and voice as they are; the avatar speaks the audio your agent already produces.
5. Run both on a slice of traffic and compare start-up time and cost per conversation before moving everyone.