How much does GPT-Live API cost? The short answer is just below: $0.05 per minute of voice session, billed per second with no rounding, plus whatever your backend model burns in tokens. I’ve dug through OpenAI’s official model documentation so you can budget a real voice agent, not a demo.
Short answer
- Voice sessions cost $0.05 per minute, billed per second — and per OpenAI’s model docs, duration is not rounded to the nearest minute.
- Backend model and tool usage are charged separately — the voice layer delegates reasoning to whatever agent you plug in behind it.
- GPT-Live-1 is only available on the Live endpoint (
v1/live/sessions) — not Chat Completions, Responses, Realtime, Assistants or Batch. - It’s a full-duplex model: it listens and speaks at the same time, with smooth interruption handling.
- API access opened in the second week of September 2026; the facts here come from OpenAI’s official model documentation, fetched 14 September 2026.
How much does GPT-Live API cost? The exact numbers
OpenAI’s official model page for GPT-Live-1 (fetched 14 September 2026) puts it in one line: “Voice sessions cost $0.05 per minute, billed per second.” Session duration is not rounded to the nearest minute, so a 30-second interaction genuinely costs $0.025, not $0.05.
That per-minute price buys you the voice layer itself — the docs describe GPT-Live-1 as “a full-duplex voice model for real-time conversations” that “can listen and speak at the same time, and delegate reasoning and tool use to a backend agent.” The word delegate is the key to your real bill: the model that does the thinking — and any tools it calls — is charged separately at that model’s own rates.
So the honest formula is:
Total cost = (session minutes × $0.05) + backend model tokens + tool usage.
This shipped to API users in the second week of September 2026 — developer access was announced on 10 September 2026 — after GPT-Live launched inside ChatGPT Voice earlier in the summer. The difference now is you can bolt OpenAI’s voice front-end onto your agent stack.
What the $0.05 per minute actually covers
From the official model page, inside the session price you get:
- Full-duplex audio — simultaneous listening and speaking, which is what makes interruptions feel natural instead of walkie-talkie awkward. OpenAI bills it as its “premier model for natural, expressive voice conversations with smooth interruption handling”.
- Streaming and function calling — supported natively.
- Delegation — the voice model hands hard reasoning and tool use to a backend agent while keeping the conversation flowing.
And what you don’t get: structured outputs, fine-tuning and predicted outputs are all unsupported, and the model lives exclusively on v1/live/sessions. If your stack assumes Chat Completions or the Realtime endpoint, this is a new integration, not a model-string swap. The documented snapshot is gpt-live-1 with a knowledge cutoff of 31 July 2025 — fine, because current knowledge is the backend agent’s job.
🔥 Want this set up without the guesswork? Voice agents are exactly the kind of build we break down inside the AI Profit Boardroom — 3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap, including how I wire agents like this into actual client-getting systems. Prefer one-on-one? Book a free SEO strategy session and we’ll scope your voice agent idea together.
How much does GPT-Live API cost at scale? Worked examples
Per-second billing makes the maths refreshingly honest. Here’s the voice-layer cost (backend tokens excluded) at different volumes:
| Scenario | Voice time | Voice-layer cost |
|---|---|---|
| Quick support triage call | 90 seconds | $0.075 |
| Discovery call screener | 10 minutes | $0.50 |
| 100 calls/day × 5 min | 500 min/day | $25/day (~$750/month) |
| Always-on receptionist, 8h/day of actual talk | 480 min/day | $24/day |
Two budgeting notes from experience. First, the backend model usually costs more than the voice layer once conversations get substantive — a chatty session that triggers lots of tool calls can double or triple the effective rate, so prototype with logging on before promising a client a number. Second, because billing is per second of session, dead air you keep open still bills — close idle sessions programmatically.
If you’re comparing ecosystems before committing, my breakdowns of GPT-6 Astra with Hermes Agent and what actually leaked about GPT-6 and Doug cover the reasoning side of the stack, and GPT Image 2.5 Sunburst vs Flare shows how OpenAI is tiering its media models — the same pattern now arriving in voice.
Where a voice agent actually makes money
A voice layer is only worth $0.05 a minute if the minutes do a job. The builds I rate: inbound lead qualification (the agent answers, qualifies and books — then a human closes), appointment-setting for local businesses, and voice front-ends on internal SEO tooling where dictating beats typing. In every one, the voice model is the cheap part and the system around it — scripts, tools, follow-up — is where the leverage lives. That system layer is what we build together in the AI Profit Boardroom.
How much does GPT-Live API cost? The bottom line
$0.05 per minute, billed per second, no rounding — plus your backend model’s tokens and tools, billed separately. It’s on the Live endpoint only (v1/live/sessions), snapshot gpt-live-1, with streaming and function calling in, and structured outputs, fine-tuning and predicted outputs out. For a natural, interruptible voice front-end on your own agent, that’s a very buildable price — just budget the whole stack, not the headline number.
Want the shortcut? Book a free SEO strategy session and I’ll help you decide whether a voice agent belongs in your funnel at all — or join the AI Profit Boardroom and build it alongside people already shipping.
FAQ: how much does GPT-Live API cost
Is there a free tier for the GPT-Live API?
OpenAI’s model documentation doesn’t list a free tier for GPT-Live-1 voice sessions. You pay $0.05 per minute, billed per second, from the first session. Standard OpenAI platform credits apply as normal.
Does the $0.05 per minute include the reasoning model?
No. Per the official docs, backend model and tool usage incur separate charges. GPT-Live-1 handles the voice conversation and delegates reasoning and tool use to a backend agent, which is billed at that model’s own token rates.
Which endpoint does GPT-Live-1 use?
Only the Live endpoint, v1/live/sessions. It is not available through Chat Completions, Responses, Realtime, Assistants or Batch.
Is billing rounded up to the nearest minute?
No — the docs state voice sessions are billed per second and session duration is not rounded to the nearest minute. A 90-second call costs $0.075.
Can GPT-Live-1 call functions and stream?
Yes, it supports streaming and function calling. It does not support structured outputs, fine-tuning or predicted outputs, per the model page.
What snapshot should I use in production?
The documented snapshot is gpt-live-1, with a knowledge cutoff of 31 July 2025. Remember its job is conversation, not knowledge — pair it with a current backend model for facts and tools.
Related reading
- GPT-6 Astra + Hermes Agent: Setup, Cost & Verdict
- GPT-6 & Doug: What Actually Leaked (2026)
- GPT Image 2.5 Sunburst vs Flare: Which One Wins?
Next step: if you’re costing a voice agent for your business, don’t do it alone. The AI Profit Boardroom has the templates and four live calls a week to pressure-test your build, or book a free SEO strategy session and we’ll run your numbers together.
About the author
Julian Goldie is an SEO agency owner with 10+ years in SEO, 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, and a best-selling SEO book. He publishes daily AI SEO tutorials on YouTube.
Want to learn AI SEO with him? Join the AI Profit Boardroom (3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap) or book a free SEO strategy session for a custom plan — and if you need done-for-you SEO, book a call with the Goldie Agency for a custom quote.
Last updated September 2026. This is the living guide to how much does GPT-Live API cost — it gets updated as the tools change.