OpenAI GPT-Live Now Available for Telnyx Voice AI Assistants

Telnyx has added OpenAI's GPT-Live full-duplex speech-to-speech model, openai/gpt-live-1, to its Voice AI Assistants, letting callers interrupt mid-sentence while the conversation continues. The model requires customers to supply their own OpenAI API key as an integration secret, with OpenAI billing the GPT-Live session and Telnyx billing the call and platform usage; call audio streams to OpenAI over a WebSocket, bypassing separate STT and TTS steps. GPT-Live assistants are limited to OpenAI Live voices (marin is the default), cannot use a fallback model, and cannot be combined with workflows, and GPT-Live cannot call tools on its own — reasoning and tool calls are handed to a delegation model so tools, MCP servers, call transfers, and hangups keep working.

Contact us https://telnyx.com/contact-us Log in https://portal.telnyx.com OpenAI GPT-Live openai/gpt-live-1 is now available in Telnyx Voice AI Assistants. GPT-Live is OpenAI's full-duplex speech-to-speech model: it listens and speaks at the same time, so callers can interrupt mid-sentence and the conversation keeps its rhythm. Telnyx provides the phone number, the call, and the tool layer around it. - OpenAI GPT-Live model: Select openai/gpt-live-1 as the model for a Voice AI Assistant in Mission Control. - Bring your own OpenAI key: GPT-Live requires your OpenAI API key, added to Telnyx as an integration secret. OpenAI bills the GPT-Live session to your OpenAI account, and your Telnyx call and platform usage stays on your Telnyx account. - Speech-to-speech calls: Call audio streams to OpenAI over a WebSocket, so the conversation itself does not pass through separate STT and TTS steps. - OpenAI Live voices: The Voice tab lists OpenAI Live voices for GPT-Live assistants, with marin as the default. - Delegation for reasoning and tools: GPT-Live handles the conversation and hands reasoning and tool calls to a delegation model. Your assistant's tools, MCP servers, call transfers, and hangups keep working through delegation. - Current limits: GPT-Live assistants use OpenAI Live voices only, cannot use a fallback model, and cannot be combined with workflows. - Direct SIP option: To run GPT-Live from your own backend instead of a Telnyx assistant, the GPT-Live SIP guide shows how to route a Telnyx number straight to OpenAI over encrypted SIP and control the session from your application. - GPT-Live handles turn-taking and barge-in inside the model, so callers can interrupt and talk over the agent without breaking the conversation. - GPT-Live cannot call tools on its own. Delegation routes that work through the Telnyx tool stack, so an assistant can look up records, transfer a call, or trigger a webhook mid-conversation. - The phone number, call routing, assistant configuration, and tools stay in one Telnyx system, with OpenAI as the model provider on your own key. - Inbound support lines where callers talk over the agent and expect it to stop and listen. - Appointment booking and scheduling calls that need natural turn-taking. - Call flows that pair a fast conversational front end with tool-heavy back-end lookups, such as order status or account changes. 1. Log in to the Telnyx Mission Control Portal https://portal.telnyx.com/ . 2. Add your OpenAI API key as an integration secret. 3. Navigate to AI → Assistants and create a new assistant or select an existing one. 4. On the Agent tab, set the model to openai/gpt-live-1 and select your OpenAI API key secret. 5. On the Voice tab, choose an OpenAI Live voice. 6. Save the assistant and place a test call. Learn more in the Voice Assistant quickstart https://developers.telnyx.com/docs/inference/ai-assistants/no-code-voice-assistant , the OpenAI GPT-Live model page https://developers.openai.com/api/docs/models/gpt-live-1 , or the Telnyx GPT-Live SIP guide https://developers.telnyx.com/docs/voice/sip-trunking/gpt-live-configuration-guide .