HeyGen API vs LiveAvatar Pricing: What a Live Conversation Costs

HeyGen video generation and LiveAvatar live conversations are different products with different billing. For a live voice-and-video application, compare LiveAvatar's FULL or Avatar Only mode rather than a generated-video price. Checked on October 12, 2026, LiveAvatar lists Essential at $99 per month with 1,100 plan credits and Business at $475 with 6,000. FULL uses two credits per session minute; Avatar Only uses one, excluding the conversation services you bring yourself. Product scope matters as much as the headline price.

Which HeyGen product does your application need?

A generated-video API returns an asset that can be rendered, downloaded or published. A live avatar service maintains a session in which a person can speak and receive responses. Those workloads have different cost drivers: generated length and resolution on one side; connection time, speech, model inference, concurrency and idle time on the other. Start by deciding which output you need before comparing a credit pack or a per-minute figure.

HeyGen's LiveAvatar documentation describes a separate real-time product and billing system. Do not assume an editor subscription, generated-video allowance or existing avatar asset automatically transfers. Check the relevant product's current documentation and agreement for compatibility. This guide focuses its numerical comparison on the live-session prices verified below; it does not repeat older HeyGen rendering or editor prices as if they were current quotes.

LiveAvatar plan credits are not the same as minutes

• Essential: $99 per month, 1,100 plan credits, one custom 720p avatar, up to 20 minutes per session, and advertised unlimited concurrency.
• Business: $475 per month, 6,000 plan credits, one custom 1080p avatar, up to 60 minutes per session, and advertised unlimited concurrency.
• The billing documentation additionally displays a +10-credit allowance beside the paid-plan quantities. The calculations below use the advertised plan credits, without treating that extra allowance as a recurring pricing assumption.
• FULL mode consumes two credits per minute. Avatar Only, also described as LITE, consumes one credit per minute. A lower Avatar Only price does not include the cost of your own conversation stack.

If every included credit is used for FULL sessions, Essential's plan credits represent 550 minutes and Business's represent 3,000. Dividing the subscription by those minutes gives about $0.18 and $0.158 per included minute respectively. In Avatar Only mode, the same plan quantities represent 1,100 and 6,000 minutes, or about $0.09 and $0.079 per included minute before your other providers. These are full-utilization calculations, not guaranteed effective prices: unused credits raise the cost of each minute you actually use.

The current billing documentation lists Essential overage at $0.095 per credit and Business at $0.09. That makes FULL overage $0.19 or $0.18 per minute; Avatar Only overage is $0.095 or $0.09. Keep included-minute arithmetic separate from overage. An application with modest traffic may spend most of its budget on the base plan, while a high-use integration may pay predominantly for overage.

• LiveAvatar official credit rates and subscriptions
• LiveAvatar current plan features

Why the enterprise headline is not your all-in quote

LiveAvatar advertises rates down to $0.01 per minute at the top enterprise volume tier. Its site labels this as a streaming rate and enterprise terms are custom. It is not the self-service FULL rate, nor proof that speech recognition, your selected language model, voice, support and reserved capacity will all be supplied for that amount. Ask for a written scope and minimum commitment before using an enterprise headline in a budget.

Likewise, advertised unlimited concurrency does not by itself establish a customer-specific reserved-capacity agreement or a measured latency guarantee. Separate admission limits, maximum session duration, actual available capacity and contractual service levels. Evaluate the workload you expect, including cold starts and failed or interrupted sessions, instead of treating the longest feature list as the lowest cost.

How Realtime Avatar's managed session pricing compares

Realtime Avatar's published Developer plan is $24 per month with 600 included realtime minutes and $0.085 per minute overage. Studio is $119 with 3,000 included minutes and $0.08 overage. Fully consuming those call allowances gives approximately $0.04 per included minute for either plan. Actual usage, unused allowance and other credit-consuming operations change the effective price. Consult the current catalog before purchasing; this is a dated comparison, not a permanent cheapest-provider claim.

RTA's managed live session includes speech recognition, the language model, speech synthesis and rendered avatar video. Your application hosting and external tools remain separate. Creating a character uses a single portrait image; new creation from a supplied video is not the general public path. The platform generates the resting loop, and additional avatar creation beyond the plan allowance is separately priced. The same credit wallet can pay for character generation and calls, so do not assume every free or included credit will become a conversation minute.

The free RTA allowance is 480 credits once per account, equivalent to 8 minutes only if all of it is spent on air. Creating an avatar consumes some of that same balance. This distinction matters when estimating the cost of onboarding a user who creates a character before making a first call. It also prevents a free-trial promise from implying more conversation time than the actual wallet can fund.

• Current RTA plans and allowances
• Create an avatar from a portrait
• Session limits and client integration

Compare the integration you will actually deploy

If you already operate your own voice and conversation stack, verify the exact input interface instead of assuming all avatar products are interchangeable. An approved external language-model endpoint is not the same as arbitrary audio passthrough. RTA's public SDK does not provide an input for live audio from your own text-to-speech provider; the managed voice pipeline remains part of the product. Unsupported integration requirements can matter more than a few cents in a price table.

For a useful evaluation, use the same conversation task and intended number of concurrent users. Record the time until a user actually sees and hears the character, whether interruptions work, how the session ends, and what the usage record bills. Then add model and voice charges not included in the selected mode, idle infrastructure, generation failures, support effort and any contractual minimums. Keep those measured results separate from vendor targets. Choose the product that completes your user's task within an honest cost and permission boundary, not the one with the smallest isolated number.