Is Gemini Live Avatar Free

No. If your question is “is Gemini Live Avatar free,” Google’s September 24, 2026 release is a metered Gemini Enterprise API feature, not an entitlement documented for the consumer Gemini app. Six usage lines can contribute to the bill, and a custom face still stops at an enterprise allowlist.

Is Gemini Live Avatar Free?

Gemini 3.8 Live with Live Avatar is not a permanent free feature. Google offers the model through Gemini Enterprise Agent Platform with Standard pay-as-you-go billing or Provisioned Throughput, while built-in avatars and custom likenesses have different access rules.

The important split is not simply free versus paid. It is consumer app versus Enterprise API, then built-in avatar versus custom likeness.

Google’s current consumer plans answer a different buying question. The Gemini pricing guide covers those plans and explains why app subscriptions and API bills must be budgeted separately.

Gemini Enterprise Live Avatar: What Changed on September 24

The September 24 change made gemini-3.8-live generally available for production use with live avatar output. Google’s GA announcement says the feature can generate a synchronized talking avatar for web, mobile, and kiosk experiences, while the model page documents Standard PayGo in us, eu, and us-central1.

The underlying product is a stateful, real-time session. It accepts text, audio, and image or video input, then can return speech and an MP4 avatar stream over a WebSocket connection. The avatar is not a prerecorded presenter pasted onto a chatbot. It is another output modality inside a live, interruptible conversation.

That distinction matters for a customer-service design. The same model can listen, see a shared camera feed, call a backend tool, and keep the interaction moving. The earlier analysis of Gemini 3.8 Live background tool calls covers the voice-agent behavior. This release adds the face and its own billable output line.

What did not change is just as important. A consumer Gemini subscription did not become an API entitlement. A custom employee or spokesperson likeness did not become open self-service access. Gemini 3.8 Live Extended Thinking also remained in private preview at launch.

What It Means for Builders, Operators, and Buyers

Builders: instrument six meters, not one session timer

Builders should start with a stock avatar and collect the usage returned by every turn. Google bills text input, image or video input, audio input, text output, audio output, and avatar-video output separately. Previous context can also be reprocessed on later turns, so two sessions of equal duration can have different bills.

The engineering consequence is simple: a stopwatch is not cost telemetry. Log billed tokens by modality, speaking time, whether the camera was enabled, turn count, and successful versus failed requests. Google says input and output are charged only when a request returns HTTP 200.

Operators: the speaking ratio is a product decision

Operators control a large cost lever through conversation design. Avatar-video output is charged only while the avatar is speaking, not while it is idle and listening. A concise agent that asks one clear question and waits creates less avatar output than an agent that repeats policies, narrates every tool call, or delivers long acknowledgements.

That is not permission to make the experience abrupt. It is a reason to design short turns, visible progress, and purposeful silence. The best operating metric is resolution quality per session, with modality costs attached, not the lowest possible token count.

Buyers: stock and custom are different procurement paths

Buyers can evaluate a built-in avatar through the documented console flow, subject to project access and billing. A custom face requires a separate approval path. If the business case depends on a founder, employee, paid performer, or brand character appearing on screen, the allowlist belongs at the start of procurement, not after the prototype wins internal approval.

The commercial decision therefore splits in two. Use a stock avatar to prove that a visible agent improves the journey. Wait for written custom access, likeness rights, security review, and measured session economics before promising a branded face in production.

Gemini Live Avatar Pricing Has Six Meters

The current non-global rates are not a subscription. They are line items per 1 million tokens. The figures below were verified against Google’s live pricing table on September 27, 2026.

The worksheet uses a declared synthetic ten-minute customer-service session: 40,000 billed text-input tokens, 8,000 text-output tokens, four minutes of caller speech, three and a half minutes of avatar speech, and a camera stream enabled for the full session. Google documents 25 audio tokens per second, 258 tokens per image, one input frame per second, and 6,192 avatar-video tokens per speaking second.

Gemini 3.8 Live Avatar Cost: A Published-Rate Worksheet

The avatar row deserves special attention. Its rate looks low at $1 per 1 million tokens, but Google converts active avatar speech at 6,192 tokens per second. In the example, 210 speaking seconds produce 1,300,320 avatar-video tokens and a $1.30032 subtotal. Idle listening does not create that video-output charge.

Audio is separate. The same 210 seconds create 5,250 audio-output tokens at the documented 25 tokens per second, adding $0.06300. The face does not replace the voice meter.

Camera input is also optional and separate. Ten minutes at one frame per second produces 600 images. At 258 tokens per image, that becomes 154,800 input tokens and $0.15480. Turn the camera off and that illustrative line disappears, but the audio, text, and avatar output lines remain.

The least predictable line is accumulated context. Google says each new turn can reprocess the current turn plus prior tokens still in the session window. A long, tool-heavy conversation can therefore increase text input even when the visible prompt looks short. Use returned usage, not a duration multiplier, for the production forecast.

Can Google Cloud Credits Make It Free?

Cloud credits can make an eligible pilot cost nothing out of pocket for a while, but they do not create a free Live Avatar tier. Google’s Free Trial documentation offers eligible new customers a $300 Welcome credit valid for 90 days.

Eligibility is narrow. You must never have been a paying user of Google Cloud, Google Maps Platform, or Firebase, and you must not have claimed the Cloud Free Trial before. Google also warns that Free Trial access can be limited to prevent abuse.

There is one naming trap. Google specifically says the $300 credit cannot pay for Gemini API costs in AI Studio. Gemini Enterprise Agent Platform is a different product surface, but the Live Avatar model page does not promise that every free-trial project will receive this model or a custom-avatar entitlement. Confirm access inside the target project before putting the credit into a launch budget.

The practical accounting rule is to show two columns: metered cost and cash paid after credits. The first reveals whether the product is viable after a promotion ends. The second tells finance what leaves the account during the pilot. Calling the product free erases the first column.

Gemini Custom Avatar Access Stops at the Allowlist

Built-in avatars are the self-serve starting point. Google’s configuration guide documents choosing a prebuilt avatar and voice in Agent Studio or naming both in the API configuration.

A custom likeness is different. Google’s GA announcement says the feature is available only through an enterprise allowlist and verification process, and directs customers to their Google Cloud account team. The documentation may show an Upload control, but a documented control is not proof that a particular project is entitled to use it.

The input itself has a craft bar. Google recommends a PNG in RGB color, at least 704 by 1280 pixels and 720p, with a file smaller than 5 MB. A 9:16 portrait is standard. Head and shoulders should occupy more than 60% of the frame against a simple background, with a level face, direct gaze, and neutral expression. The safety guidance excludes minors and celebrities.

This is still a raster likeness, meaning a pixel-based image rather than an editable vector portrait. Clean source pixels help the model, but they do not establish usage rights. Google places responsibility for face and voice consents on the customer, and says generated audio and video carry imperceptible SynthID watermarks.

The same service brief, before and after

The first instruction in Google’s SDK example is broad: “You are a helpful customer service assistant.” That is enough to prove a connection, not enough to judge a customer-facing experience.

A safer synthetic pilot brief is more constrained:

You are a support guide for a fictional outdoor retailer. Ask for the order number, explain that no real account lookup is connected, summarize the issue, and escalate payment or safety complaints to a human. Never claim a refund was issued.

This is an instruction rewrite, not a reported test result. Its job is to make the pilot inspectable: the agent must disclose the missing integration, avoid fake actions, and expose the handoff moment an operator needs to review.

What to Do Differently Now

Act now if a stock avatar is acceptable and the team can meter a contained proof of concept. Wait if the product promise requires a custom spokesperson, a guaranteed fixed cost, or an Extended Thinking feature that remains private. You are unaffected if the need is ordinary consumer Gemini Live or scripted narration rather than a real-time visual agent. For the latter, use the separate Gemini 3.8 TTS pricing guide.

The decision rule is direct: prove the interaction with stock media, prove the bill with usage records, then request the custom face. Reversing that order creates a polished demo whose most important production dependency is still unknown.

The Monday move

1. 1

   Set the pilot boundary

   Choose one synthetic support journey, one maximum session length, and one escalation rule. Decide in advance whether the camera is necessary, because live image input creates its own meter.

2. 2

   Open a stock-avatar session

   In Agent Platform Studio, open Stream realtime, switch to gemini-3.8-live, select Live Avatar, then choose a built-in avatar and prebuilt voice. Add system instructions before starting the session.

3. 3

   Use the constrained brief

   Paste the fictional-retailer instruction above. Use synthetic order details and no personal customer data. The purpose is to observe flow and capture usage, not to simulate a completed refund or a connected backend.

4. 4

   Reconcile all six lines

   Save the billed usage by modality and compare it with Cloud billing records. Record avatar speaking seconds, camera state, turn count, context growth, and failed requests. Do not divide the total by minutes and call that a universal rate.

5. 5

   Test the custom gate last

   Only after the stock pilot earns a business case should the buyer ask the Google Cloud account team for custom-avatar access. If the project is not allowlisted, stop the branded-likeness plan rather than treating the restriction as an implementation bug.

What Is Overhyped

“Generally available” does not mean every part is broadly self-serve. The model and stock-avatar path are GA, while Google still describes custom avatars as allowlisted.

Free consumer Gemini Live does not make Enterprise Live Avatar free. The similar names hide different products, contracts, and meters.

The $1 avatar-video rate is not a $1 session. It is a price per 1 million avatar-video tokens, and Google converts active speech at 6,192 tokens per second. Five other lines can bill alongside it.

A fixed per-minute headline is too neat for this API. Speaking ratios, camera use, output mix, context accumulation, and tool-heavy turn structure all change the result.

A face does not prove a better service experience. Video earns its cost only when it improves comprehension, trust, accessibility, or completion for a named journey. If a voice-only agent resolves the same task, the avatar is presentation spend without an operating gain.

Gemini Live Avatar FAQ

Can you use Gemini Live for free?

Consumer Gemini Live follows separate app terms, while Gemini 3.8 Live with Live Avatar in Gemini Enterprise is metered. Do not use the consumer answer to budget the Enterprise API.

How to enable avatar in Gemini?

In Gemini Enterprise Agent Platform, open Studio > Stream realtime, choose gemini-3.8-live, select Live Avatar, then choose a built-in avatar and voice. A custom likeness requires allowlist approval.

How much is a Gemini Live subscription?

Google does not document a flat Live Avatar subscription. The Enterprise API charges six input and output meters; consumer Gemini subscriptions are separate.

Is Gemini completely free to use?

No. Google has separate product allowances and temporary Cloud credits, but the Enterprise Live Avatar release has published pay-as-you-go rates.

How long can I use Gemini for free?

Eligible new Google Cloud customers receive a $300 Welcome credit valid for 90 days. That is a temporary account credit, not a promised duration of free Live Avatar access.

Is Gemini better than Chatgpt now?

Live Avatar does not settle the general assistant comparison. It is an enterprise voice-and-video capability, so judge it against the customer journey, access requirements, and measured bill.

Who is better than Gemini?

There is no universal winner. For a talking avatar, compare likeness access, interaction quality, tool integration, consent controls, latency, and the full measured session cost.

What can I replace Gemini with?

A dedicated avatar platform or a separate voice-and-video stack can replace it. Compare the complete workflow and approval path, not one headline token rate.

Is Gemini still safe?

Google adds SynthID to generated audio and video and gates custom likenesses, but deployers still own consent, privacy, data governance, misuse prevention, and human escalation.

Planning a customer-facing avatar that must connect to real support workflows? See the AI customer service build path.