Skip to content

OpenAI

GPT-Live-1.1 hits the API with a cheaper standby

2 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: OpenAI models and products →
Image: threads.com

Developers can now use GPT-Live-1.1 in OpenAI’s API. GPT-Live-1.1 is OpenAI’s voice model that can listen while it speaks. The model’s page in OpenAI’s API docs lists it as gpt-live-1.1 and sets a price per minute of conversation: $0.05 while the model is active and a new, lower standby rate of $0.02.

ChatGPT users already have the model. In October GPT-Live-1.1 took over Voice in ChatGPT, Codex and dots, and at the time OpenAI said API access was “coming soon.” The model it replaces, GPT-Live-1, opened to API developers on September 10.

What’s new for developers

Older voice assistants waited for you to stop talking before they answered. GPT-Live doesn’t wait: it hears you while it speaks, so a caller can cut in mid-sentence. According to OpenAI, version 1.1 transcribes speech more accurately than GPT-Live-1 and keeps a more consistent accent and voice across languages.

The voice model doesn’t do the heavy work itself. Developers describe in plain language when it should hand a task to their backend for reasoning or tool use, and the voice model keeps talking while that runs. Testing Catalog reports that the Live tab on the OpenAI Platform lets developers try all of these features right away. In a screenshot it published, the tab has separate settings for the voice model and for the model it hands work to.

What it costs

Voice sessions are billed by how long they last. The $0.02 standby rate is meant to cut the cost of idle time, but it applies only to WebRTC sessions (a standard way to stream live audio) while inference is paused. Any other session is billed at the active rate. The per-minute price also isn’t the whole bill: the backend models and tools the voice agent hands work to are billed separately.

The limits

The model takes and returns audio and text. It doesn’t handle images or video. It supports function calling but not structured outputs or fine-tuning. GPT-Live-1.1 and GPT-Live-1 have separate caps on how many sessions can run at once, and each cap counts sessions across all inference regions.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.