5 cents a minute: what an interruptible AI host costs per webinar
By Alan Magliocca · November 13, 2026
Quick answer: OpenAI put GPT-Live-1, a voice model that can be interrupted mid-sentence, in its API on September 10, 2026 at $0.05 per minute, billed per second. That makes one 60-minute voice session $3.00, but it's the voice model's price only, so budget separately for the backend model, avatar video, streaming and the event platform.
- GPT-Live-1 costs $0.05 per minute, billed per second, with no rounding up to the next whole minute.
- One voice session running a full 60-minute event costs 60 x $0.05 = $3.00.
- That is the voice model only: backend model calls, tools, avatar video, streaming and the webinar platform cost extra.
- Cost scales with sessions, so giving every attendee a private voice conversation multiplies the bill and runs into concurrency limits.
What happened
On September 10, 2026, OpenAI made GPT-Live-1 available in its API. It is a full-duplex voice model, which means it listens and speaks at the same time and can be interrupted mid-sentence. Heavier reasoning and tool calls are handed to a backend model.
OpenAI's GPT-Live-1 model page states the price plainly: "Voice sessions cost $0.05 per minute, billed per second." It adds that session time isn't rounded up to the next whole minute, and that backend calls use the normal pricing for whichever model and tools you configure. The page lists audio and text as inputs and outputs. Image and video are not supported.
OpenAI also reported turn-taking latency of about 0.8 seconds, down from 1.41 seconds for the previous model, as AI Weekly reported. That figure comes from OpenAI's own testing. The same day, HeyGen released an open-source reference integration that connects GPT-Live-1 to a HeyGen LiveAvatar, which needs a LiveAvatar API key as well as an OpenAI key.
What it means for people selling with webinars
An AI host that an attendee can interrupt with a spoken question is no longer expensive at the voice layer. A full hour of the voice model costs three dollars. That's the headline, and it's real.
It is also only one line on the bill. GPT-Live-1 produces audio and text. It does not produce a face, a slide, a registration page or a reminder. Anything that looks like a presenter on screen comes from an avatar service. HeyGen's own reference build pairs the voice model with LiveAvatar, which is a separate account and key. When a vendor quotes you a price for an "AI host", ask which of these pieces it includes.
The other thing that drives cost is how many sessions you run. One session is one conversation. If the AI host talks to the whole room through one session, the cost per event is small. If every attendee gets their own private, interruptible conversation, every attendee is a session, and the math changes fast.
Cost math: what a 60-minute event costs in voice
All figures below are for the GPT-Live-1 voice model only, at OpenAI's listed $0.05 per minute.
One shared session for a 60-minute event:
- 60 minutes x $0.05 = $3.00
Per-second billing on an uneven session:
- $0.05 per minute / 60 = about $0.00083 per second.
- A session that ends at 47 minutes 30 seconds: 47.5 x $0.05 = $2.375, about $2.38. You don't pay for the full 48th minute.
A longer event with Q&A:
- 90 minutes x $0.05 = $4.50
An evergreen event that runs once a day for 30 days:
- 30 sessions x $3.00 = $90.00 for the month in voice.
Private conversations for each attendee (an example to show scale, not an OpenAI figure):
- 100 attendees, each with a private voice session for the full 60 minutes: 100 x $3.00 = $300.00.
- 100 attendees, each with a private 5-minute voice Q&A after the talk: 100 x 5 = 500 minutes, and 500 x $0.05 = $25.00.
Those last two lines show why design matters more than the per-minute rate. The same 100 people cost $3.00 or $300.00 in voice, depending on whether they share one host or each get their own.
There is also a ceiling. OpenAI's model page measures rate limits in concurrent sessions: 50, 300 or 500 depending on your account tier, and the free tier isn't supported. If 100 attendees each open a private session at the same moment, a 50-session limit won't cover it.
What the five cents doesn't cover
To be clear, $0.05 a minute is the price of the voice model, not the price of running an AI-hosted webinar. A full budget also includes:
- Backend model calls. When GPT-Live-1 hands off reasoning or a tool call, that call is billed at the backend model's normal rates.
- Avatar video. GPT-Live-1 doesn't output video. A talking presenter comes from a separate avatar service with its own pricing.
- Streaming to the audience. Getting audio and video to many viewers at once is its own cost.
- The event platform. Registration pages, reminders, chat, replays and sales tracking.
- Build and upkeep. HeyGen describes its reference repos as starters, not finished deployments, so someone has to connect, test and maintain the pieces.
Response speed matters too. If you want to know why 0.8 seconds is the number vendors compete on, see 0.6 vs 0.8 seconds. For a second voice supplier, see Eleven Realtime.
What to do this week
- Write down three numbers for your next event: its length in minutes, expected attendees, and whether Q&A is shared or private.
- Run the voice math: minutes x $0.05 x number of sessions. Do it for both the shared and the private design.
- Ask any AI-host vendor which voice model they use, and whether their price includes voice, backend model calls, avatar video and streaming.
- Ask how many attendees can talk to the host at the same time, and what happens when that limit is reached.
- Before you commit, hold a 10-minute internal call with the setup and interrupt it on purpose. Note how quickly it stops and answers.
- Write the line that tells attendees the host is AI, and put it at the start of the event.
How BrightStage AI helps
BrightStage AI doesn't offer a real-time voice host, so this math is for judging other vendors' quotes, not BrightStage pricing. What it does today: during live and automated events, a chat assistant answers attendees in the event chat, and a video creator builds a script and slides with AI presenters. If you're weighing a voice host against a chat-based one, the open-source interruptible presenter shows what the voice route involves. Agencies looking at the same model for a different job can read AIVX Labs' guide to building an AI receptionist with GPT-Live-1.
Sources
Frequently asked questions
How much does GPT-Live-1 cost for a one-hour webinar?
One voice session running for 60 minutes costs 60 x $0.05 = $3.00. That is the voice model only, before backend model calls, avatar video, streaming and your event platform.
Am I charged for a full minute if the session ends early?
No. OpenAI bills GPT-Live-1 per second and says session time isn't rounded up to the next whole minute.
Can GPT-Live-1 show a presenter on screen?
No. Its model page lists audio and text as outputs and says image and video aren't supported. A visible presenter needs a separate avatar service, as in HeyGen's LiveAvatar reference build.
Related reading
Written by Alan Magliocca, founder of BrightStage AI and AIVX Labs (Podavinci LLC). Written with AI assistance and checked against the linked sources. Follow along: LinkedIn · YouTube · TikTok · X · Facebook · Instagram · Skool community
