Realtime avatar API pricing: included minutes, overage, and total cost

A realtime avatar budget has three separate questions: what you pay to start the month, how much usage that includes, and what the next minute costs. A headline per-minute figure answers only one of them. Here is the calculation for Realtime Avatar, with an interactive cost calculator you can adjust to your application.

Our published self-serve overage is $0.08–$0.095 per minute, or $4.80–$5.70 per hour. The live-call product includes speech recognition, the language model, voice and avatar rendering. Your own hosting and external tools remain separate.

Current plans, with the units kept separate

These figures come from the same billing catalog as the pricing page. Sandbox is free and provides up to 17 minutes if all of its credits go to calls; it has no automatic overage. Scale is quoted individually.

Included-minute cost is different from overage

Divide a subscription by its allowance only when you are modeling full use of that allowance. For example, the Developer subscription buys an allowance whether you use it all or not. The effective rate gets higher when the allowance goes unused. Overage is the marginal rate after the allowance, not the rate you pay for every minute from zero.

monthly estimate = subscription + max(0, call minutes - included minutes) × overage rate

This model assumes monthly billing, one simultaneous call, and all included credits spent on calls. It excludes taxes, annual discounts, negotiated rates, optional credit top-ups, avatar creation, other generation, and external hosting. It does not inspect your account balance or predict a bill.

Worked usage scenarios

These are calculated scenarios, not measurements of live session performance. The lowest eligible self-serve plan is selected using the formula above. The inputs and results are reproducible in the calculator.

Now change peak concurrency to three or ten. A cheaper plan with too few simultaneous sessions becomes ineligible even when it includes enough monthly minutes. A plan limit describes eligibility; it does not guarantee that every capacity request will succeed.

Creation and idle time belong in the budget

Creating a photo avatar beyond the plan's included count has a published price of $1. Portrait generation and idle-clip generation are additional uses of the shared credit wallet. Credits spent on generation are no longer available for calls. Using a ready example avatar is the shortest way to evaluate the call itself.

Budget for connected time, including pauses. Do not estimate a ten-minute conversation as two minutes merely because the avatar spoke for two minutes. End the call explicitly when the user is finished and examine the session usage record.

Comparing LiveAvatar, Tavus, or a renderer

Use the same worksheet for each vendor: base subscription, included minutes, overage, concurrency, avatar creation, minimum session charge, and which parts of the voice pipeline you supply. A renderer-only rate needs the cost of speech recognition, an LLM and TTS added before comparison with a complete conversation stack.

Check LiveAvatar pricing and Tavus pricing at the time of your evaluation. Their consumer plans, developer plans and rendering modes are different offers. We avoid turning an unverified competitor rate into a fixed calculator input. Our architecture comparison explains the differences, and the developer evaluation checklist gives you a repeatable prototype to compare.

Try the calculation, then the product

Enter your minutes and concurrency, then follow the Next.js guide to run a call on an existing character. Compare the session's recorded usage with your assumptions before extrapolating a production budget.