OpenAI launches outbound phone calling API for $0.05 per minute

Outbound calling goes live

OpenAI has officially announced the general availability of its GPT-Live outbound telephony API. This update allows developers to connect voice agents to real phone numbers so the AI can initiate calls independently. As reported by Crypto Briefing, the company is moving toward a model where the AI can proactively reach out to people.

The rollout follows a rapid development timeline for the GPT-Live technology. The voice model first launched globally in ChatGPT on July 8, 2026, and reached the API just two months later on September 10, 2026. This latest update adds the ability to place calls on its own, completing a fast transition from a chat feature to a functional telephony tool.

The pricing for this specific API is set at $0.05 per minute. While this provides a clear baseline for businesses to model their costs, developers must also account for separate carrier charges. The total cost of a call will be the sum of the OpenAI fee and the fees from the chosen telephony provider.

How the technology connects

The core of this new functionality is outbound calling over SIP, which stands for Session Initiation Protocol. SIP acts as a switchboard operator for internet-based phone calls by setting up the connection and managing the session. By using SIP, OpenAI avoids having to build the complex infrastructure required to reach the public phone network.

Developers do not need to build this plumbing from scratch because OpenAI has partnered with established providers. Named launch partners include Telnyx, Twilio, and Ozonetel. These external providers handle PSTN termination, which is the technical process of carrying the call onto the traditional public switched telephone network.

OpenAI maintains control over the intelligence side of the conversation, including audio processing and session logic. The new telephony path joins existing methods like WebRTC and WebSocket transports. This means the AI stays in charge of the 'brain' of the call while the partners handle the actual wires and signals.

High-fidelity and menu navigation

Two specific technical additions make these calls feel more like human interactions. The first is HD Voice support, which provides wideband, higher-fidelity audio. This ensures that the voice quality remains clear and natural rather than sounding like a standard, low-quality phone line.

The second addition is bidirectional DTMF signaling. DTMF stands for dual-tone multi-frequency, which are the beeps a phone makes when a user taps a number on the keypad. Because the support is bidirectional, the AI agent can both send and receive these tones during a live call.

This capability is particularly useful for navigating Interactive Voice Response (IVR) systems. These are the automated phone menus that often ask users to press a specific number for billing or support. A GPT-Live agent can now listen to those prompts and press the correct buttons to navigate the system on behalf of a user.

Full-duplex human conversation

A key feature of the GPT-Live voice agents is that the calls are full-duplex. This means that both the AI and the human can talk and listen at the same time, much like a normal human conversation. It removes the awkward pauses often found in older text-to-speech systems that had to wait for a person to finish before responding.

The goal of this design is to eliminate what some have called "press 1" paralysis. By allowing the AI to handle the navigation and the conversation fluidly, it can interact with complex systems more naturally. This makes the AI a more effective tool for handling high-volume tasks that require real-time interaction.

Businesses that already use providers like Telnyx or Twilio can plug GPT-Live into their existing setups. They do not need to rebuild their entire telephony stack to use this new feature. This ease of integration is intended to help teams move quickly from testing a prototype to deploying a functional voice agent.

The role of the providers

OpenAI is making a strategic choice by not trying to become a phone carrier itself. By leaving the network side to established providers, they can focus on the intelligence layer of the technology. This separation of concerns allows the AI to remain the primary focus while the infrastructure remains reliable and regulated.

The providers handle the heavy lifting of connecting to the public switched telephone network. This means they manage the regulatory requirements and the physical routing of the calls. OpenAI simply provides the voice and the logic that determines what the agent says and how it reacts to the caller.

For developers, this means a modular approach to building AI products. They can choose the best carrier for their specific region or needs while using OpenAI's consistent API for the voice interaction. This flexibility is designed to support a wide range of use cases, from simple notifications to complex customer service interactions.