GPT-Live-1 API: Pricing, Features and How It Works

01

Overview

OpenAI released GPT-Live-1 in the API on September 10, 2026, turning the full-duplex voice technology already used in ChatGPT Voice into a developer-facing building block for apps, phone agents and business workflows.

02

What is GPT-Live-1?

GPT-Live-1 is a voice model designed for continuous conversation. It can listen while it speaks instead of forcing every interaction into strict user-turn/assistant-turn boundaries. That matters for interruptions, short acknowledgements, hesitation, background speech and other behavior that makes real conversations difficult for conventional voice-agent pipelines.

OpenAI separates the live conversation layer from deeper reasoning. GPT-Live-1 can handle the spoken interaction while delegating reasoning or tool calls to a backend text model selected by the developer. This means GPT-Live-1 should not be understood as a replacement for every model behind an agent. It is the real-time voice layer that can work with a reasoning model, tools and an agent harness.

03

What changed with the September API launch?

GPT-Live itself debuted in ChatGPT Voice on July 8, 2026. The September 10 announcement is a separate material event because GPT-Live-1 is now available directly through the OpenAI API for developers.

OpenAI highlights several capabilities for the API release: simultaneous listening and speaking, smoother interruption handling, prompt-controlled tone and pace, better handling of silence and background noise, improved long-session reliability, native ASR transcripts and response text, keyword biasing, turn detection, and telephony support.

04

GPT-Live-1 pricing

OpenAI currently lists GPT-Live-1 at $0.05 per minute for the front-end voice layer.

That is not necessarily the full cost of a production voice agent. A developer may also pay for the backend reasoning model, tools or external APIs, agent infrastructure and any telephony provider used to place or receive phone calls. A useful cost estimate therefore needs to separate the GPT-Live-1 voice-layer charge from the rest of the stack.

Custom voice access is not presented as a self-serve feature in the launch announcement; OpenAI directs interested customers to contact sales.

05

How GPT-Live-1 works

Traditional voice agents often chain speech-to-text, a language model and text-to-speech. GPT-Live-1 handles listening and speaking inside one continuous voice layer. It can respond to an interruption or acknowledgement while deeper work is delegated to another model.

OpenAI says developers can choose the backend model and tools that fit a task. A high-volume scheduling workflow might prioritize speed and cost, while a difficult support request could delegate to a stronger reasoning model. OpenAI also published an example in which conversation context is passed to Codex and its answer is returned to GPT-Live-1; the example intentionally omits connection setup and delegation handling, so it should not be treated as a complete integration tutorial.

06

Connection and telephony options

OpenAI's Live documentation organizes implementation around real-time connections including WebRTC, WebSockets and telephony/SIP. The right transport depends on where the agent runs: browser and client experiences often favor WebRTC, server-side applications can use WebSockets, and phone-agent workflows can use the telephony/SIP path.

07

Where GPT-Live-1 may be useful

The clearest use cases are experiences where conversational timing matters: customer support, restaurant reservations and ordering, appointment workflows, tutoring and language practice, hands-free assistants, and voice interfaces for agentic or coding systems.

For agencies and business teams, the important change is that a natural voice interface can sit in front of existing tools and reasoning systems instead of requiring the entire business workflow to live inside one voice model.

08

Benchmarks: what OpenAI claims

OpenAI reports that GPT-Live-1 improves Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1. It also says a GPT-Live-1 system paired with GPT-6 Astra at medium reasoning effort ranks first on its cited Tau3 evaluation.

These are first-party OpenAI evaluation claims. They are useful evidence about what the company optimized for, but they should not be treated as independent benchmark validation until third parties reproduce comparable tests.

09

Important limitations

The $0.05/minute figure covers the front-end voice layer, not every component in an agent. Exact GPT-Live-1 rate limits were not established from the launch announcement. Custom voice access is gated through sales. OpenAI says voice and language options will continue expanding, which means availability details can change. Production teams should also validate latency, transcription accuracy, interruption behavior, tool-call reliability, privacy requirements and total operating cost in their own environment.

10

FAQ

Is GPT-Live-1 available in the API?

Yes. OpenAI announced API availability on September 10, 2026.

How much does GPT-Live-1 cost?

OpenAI lists the front-end voice layer at $0.05 per minute. Backend models, tools, infrastructure and telephony can add separate costs.

Can GPT-Live-1 handle interruptions?

Yes. Full-duplex interruption handling is one of the central capabilities highlighted by OpenAI.

Can GPT-Live-1 call tools?

GPT-Live-1 can delegate deeper reasoning and tool use to a paired backend model and agent system. The exact architecture is controlled by the developer.

Does GPT-Live-1 work for phone agents?

OpenAI explicitly documents telephony support, and its Live documentation includes telephony/SIP connection paths.

Is GPT-Live-1 the same launch as GPT-Live in ChatGPT?

No. GPT-Live was introduced in ChatGPT Voice in July 2026. September 10 marks the materially new developer/API availability event.

11

Sources

OpenAI's September 10 GPT-Live-1 API announcement, OpenAI Live developer documentation, and the July GPT-Live launch announcement are the primary sources for this article.