GPT-Live-1 pricing: what the $0.05/min API really costs

Why I care what a voice minute costs

I build the integrations and API plumbing at eesel, and we have spent the last three-plus years putting AI agents on live support queues. The pattern I have watched play out over and over is that a voice or chat model looks almost free at the sticker price, and then the real cost shows up in the parts nobody quoted: the reasoning model behind it, the retries, the escalations, the long calls that never resolve. So when OpenAI put a clean $0.05/min number on GPT-Live-1, my first instinct was not "cheap," it was "what is that number not counting?"

This post answers exactly that. I will walk the one number OpenAI published, the two cost layers it hides, the ChatGPT plans that bundle the model for free, and what all of it means if you are pricing a voice agent for customer support.

The one number: $0.05 per minute

Here is the quotable line, straight from OpenAI's launch: GPT-Live-1 is available in the API at $0.05 per minute for the front-end voice layer. It went live for developers on September 10, 2026, after first shipping inside ChatGPT in July.

What that flat per-minute rate buys you is the thing GPT-Live-1 is actually good at: a single model that reasons over incoming and outgoing audio together, so it can be interrupted mid-sentence, stay quiet during a pause, and keep talking while work happens in the background. It replaces the old stitched-together speech-to-text plus model plus text-to-speech pipeline, the kind you would price by comparing something like ElevenLabs pricing against a separate reasoning model, which is why one healthcare developer, Tony Stoyanov, said it let them cut 23,000 lines of code and simplify their codebase by 80%.

Flat per-minute pricing is also a real simplification over the old realtime models, where you metered audio input and output tokens separately and had to model caching to predict a bill. One number per minute is easier to forecast. It is just not the only number.

The two layers the $0.05 doesn't cover

GPT-Live-1 is deliberately a voice layer, not a brain. When a question needs search, reasoning, or a tool call, it delegates to a separate backend model like GPT-6 Astra or a third-party model, and keeps the conversation flowing while that work runs. That architecture is great for latency. It also means your bill has three parts.

1. The voice layer: the flat $0.05/min for GPT-Live-1 itself.
2. The backend model tokens: whatever reasoning model you pair it with. OpenAI's own guidance is to route light, high-volume tasks like scheduling to a cheap model like Luna, and hard, reasoning-heavy issues to a heavier model like Astra. That token cost is entirely on you and varies with how much the model thinks.
3. The agent harness and tools: the framework, the tool calls, the retrieval, the ticketing system writes. This is your infrastructure cost, not OpenAI's line item, but it is real.

So the honest way to read GPT-Live-1 pricing is: $0.05/min is the floor. A conversation that stays chatty and rarely delegates hugs that floor. A conversation that constantly kicks hard questions to a heavy backend model can cost a multiple of it. If you have ever compared AI agent cost against a human agent, this is the same lesson: the sticker price and the loaded cost are different animals.

Estimate your real per-minute bill

Rather than hand-wave it, plug in your own numbers. This estimator stacks the flat voice rate on top of a rough backend-token cost so you can see the loaded per-minute figure and a monthly total.

The point the calculator makes fast: at a medium backend weight, your loaded cost is already well above the headline $0.05. That is not a knock on OpenAI, it is just how a delegating architecture prices out. Model this before you commit a voice deployment to a budget.

GPT-Live-1 pricing inside ChatGPT

If you are not building on the API, the pricing story flips and gets simpler. In ChatGPT, GPT-Live-1 powers the voice experience and there is no per-minute meter at all, it is bundled into whatever plan you already pay for.

The plan split is straightforward:

• Free: default model is GPT-Live-1 mini.
• Go, Plus, and Pro: default model is the full GPT-Live-1.

It rolled out globally across iOS, Android, and ChatGPT.com, so the model reached most users the day it shipped. For a consumer, the practical takeaway is that the better voice model is simply the default on a paid tier, and even Free gets a capable mini version. The metered $0.05/min world is only the developer's world.

What GPT-Live-1 pricing means for customer support

Here is where I put my support hat back on, because voice-for-support is where the per-minute model gets interesting, and a little dangerous.

Per-minute billing charges you for talk time, not for outcomes. That is fine for a language tutor or a reservation bot, and it fits plenty of conversational AI use cases. For support it inverts the incentive you actually want. A frustrated customer on a ten-minute call that ends in an escalation costs you more than a calm one resolved in ninety seconds, even though the second interaction is the win. If your resolution rate is what you care about, a meter that runs on minutes is measuring the wrong thing, and it muddies the support metrics you report on.

This is the same trap I see with a raw AI call center agent build: the voice layer is impressive and cheap-looking, and then the first-contact resolution work, the knowledge retrieval, the escalation handling, and the backend reasoning all have to be built and paid for before a single ticket is actually closed. GPT-Live-1 makes the conversation feel human. It does not, on its own, know your refund policy or your help center.

That gap is worth naming plainly, and it is not a criticism of GPT-Live-1: OpenAI built a voice layer and priced it as one. The intelligence and the business context are meant to come from the model and harness you bring.

Where eesel fits: infrastructure versus the teammate

The clean way to think about it is that GPT-Live-1 is infrastructure, and eesel is the employee. GPT-Live-1 is a model you wire into your own product. eesel is a platform where you hire a ready-to-work AI teammate for a specific job, and it arrives already carrying the skills, integrations, and company context for that role.

For support, that teammate is the AI helpdesk teammate. It joins your existing queue, learns from your past tickets and help center, and, crucially for this post, it is billed per resolution rather than per minute of talk. That means your cost lines up with tickets actually closed, not with how long a conversation drags. You can point the same teammate at a voice front end like GPT-Live-1 when you want natural phone conversations, and still keep the resolution economics sane.

And because a lot of teams reaching for a per-minute voice model are really trying to automate work programmatically, it is worth knowing eesel exposes a real CLI alongside the dashboard. It is the same AI teammate and the same workspace, driven from a terminal instead of a browser: a person can run it by hand, scripts can automate it, and coding agents like Claude Code, Codex, and Cursor can operate it directly, the same way you would drive any other ChatGPT agents workflow. So if your reason for going to the API in the first place is headless, scriptable control, you do not have to choose between "programmable" and "already knows my support context."

If you want the honest cost comparison for your own volume, the cost savings breakdown and our build vs buy guide both walk the per-minute-versus-per-resolution math in more depth.

GPT-Live-1 pricing, in one line

The number to remember is $0.05 per minute in the API, free inside every paid ChatGPT plan, with a backend model and harness stacked on top that you pay for separately. It is a fair, simple price for a much better voice layer. Just budget the whole stack, not the headline, and if the job is support, ask whether you are paying for minutes or for resolved tickets.

Want a support teammate that already knows your help center and bills per resolution, not per minute? eesel plugs into your existing helpdesk in minutes, simulates on your own past tickets before it ever goes live, and is free to try. Try eesel or book a demo to see it on your own data first.