bitHuman in 2026: Pricing, Device vs Cloud, and When to Choose It

bitHuman bills in credits, and 100 credits cost $1. As of 30 October 2026, its two current models (Essence 2 for a photoreal person, Expression 2 for any character) use 2 credits a minute on the user's device or your servers and 4 credits a minute in bitHuman's cloud: $0.02 and $0.04 at the top-up rate. Plans lower that rate if you use their credits. Creator is $20 a month for 2,000 credits, Pro $99 for 12,000, Business $299 for 50,000 and Enterprise $999 for 250,000. A device minute works out to about $0.017 on Pro, $0.012 on Business and $0.008 on Enterprise; a cloud minute to $0.033, $0.024 and $0.016. Plan credits refill each billing period and don't carry over, and top-ups are spent after them. Yearly billing saves up to 16%. If bitHuman runs the conversation too, its managed voice chat costs 10 credits a minute, or 30 with the camera on.

Where the avatar renders changes both the rate and the minutes that count. In bitHuman's cloud, a session bills its active time, talking or idle, by the exact second. On your own devices or servers, current versions of the Essence 2 and Expression 2 SDKs, the Python SDK and the CLI bill only the seconds the character speaks. Take a 10-minute support call in which the avatar talks for 4 minutes. At the top-up rate, bitHuman charges about $0.08 for it on the device and $0.40 in its cloud. Realistic Avatars bills session time per second, listening included, so the same call costs $0.20. The device rate has a catch: the device does the work. Rendering on the user's phone or browser costs you nothing, but rendering on your own servers means paying for them. bitHuman measures one Intel Core i7-13700F with no GPU rendering Essence 2 at 1.2× to 2.0× real time, depending on the SDK.

At 1,000 minutes a month with your own voice agent, bitHuman's cloud needs 4,000 credits: Creator's 2,000 plus 2,000 in top-ups, or $40. On the device, 1,000 minutes need at most 2,000 credits, which Creator covers for $20, with room to spare when the avatar spends time listening. Realistic Avatars costs $20, hosted. At 10,000 minutes, bitHuman's cloud fits in Business: $299 for 50,000 credits. On the device, Pro's 12,000 credits cover 6,000 speaking minutes, so the month costs $99 if the avatar talks 60% of the time or less, and $179 (Pro plus 8,000 top-up credits) if it never stops. Realistic Avatars costs $200. Below 1,000 minutes the plan fee decides it: 100 minutes cost $2 with us and $20 on bitHuman, because its API needs Creator. Past the $999 plan both sides quote: bitHuman's Custom plan, and our enterprise and high-volume contracts, which go as low as $0.005 a minute.

bitHuman's cheapest minute needs capable hardware. Essence 2 renders on iPhone and iPad with iOS 26, Apple silicon Macs (M3 or newer on macOS 26 for the Swift package), physical Android arm64 phones, a WebGPU browser that passes a device check, and Linux or Windows 11 PCs with or without a GPU. Each Essence 2 avatar is a file of about 140–160 MB that your app downloads. Online sessions check your credential when they start and keep rendering through a network drop of up to 5 minutes. For your own video UI on a web page, bitHuman's docs point to a cloud-rendered avatar over LiveKit, and its LiveKit plugin is Python only. Realistic Avatars renders on our servers and streams the video to your page over WebRTC or into your LiveKit room. The visitor's device only plays video, so an old laptop or a budget phone gets the same avatar as a new Mac.

bitHuman makes custom avatars cheaply. One portrait becomes an Essence 2 avatar in a few minutes for a one-time 500 credits, about $5. Expression 2 costs 2,000 credits and animates any character, from a person to an animal, a mascot or an object. Plans hold 7 to 1,000 avatars. Realistic Avatars offers a growing library of ready-made photorealistic avatars, and a custom avatar costs $199 one time, with the presenter's consent. Our avatar starts speaking less than 250 ms after your agent's audio reaches it, measured on our servers. bitHuman publishes render speed per device instead of one latency figure.

If your own agent drives a bitHuman avatar rendered in its cloud (through the LiveKit plugin or the API), switching changes only the face. If you depend on on-device rendering or the offline license, bitHuman is the one to keep.

1. List what each bitHuman agent does: its model (Essence 2 or Expression 2), where it renders, and whether bitHuman's managed voice chat runs the conversation.
2. If managed voice chat runs it, move the prompt and voice to your own stack (OpenAI Realtime, LiveKit Agents or Pipecat), or rebuild it in Voice Agents if you want it hosted.
3. Create a Talking Avatar Realistic Avatars account (it comes with $1 of free credits), create an API key and pick a ready-made avatar, or order a custom one for $199, one time, with the presenter's consent.
4. Replace the bitHuman AvatarSession in your LiveKit worker, or the bitHuman call in your server, with a Talking Avatar session that takes your agent's audio over WebRTC or in your LiveKit room.
5. Run both on a slice of traffic and compare the cost per conversation, counting the idle minutes bitHuman's cloud bills, before you cancel the plan.