GPT Live Transcribe

Audio modelCard summarized

Low-latency speech-to-text model for realtime transcription

Source: GPT Live Transcribe model page and realtime transcription guide (OpenAI API docs); no system card published

Use it via API

$0.017 / min

Code example Compare

Where to call it

OpenAI APIgpt-live-transcribe

Supports

Streaming

Full feature support

Good for

Text from microphone, call or other live streams without a spoken reply
Live captions with low-latency partial text
Domain terms and code-switching via prompt, keywords and languages

Watch out for

No timestamps, speaker labels or confidence
You must detect turns
Latency vs accuracy trade-off

Overview

OpenAI · 02Transcribe

Live Transcribe

Low-latency speech-to-text model for realtime transcription

Typical price

$0.017 / min

Realtime transcription

Max length: —
Languages: —
Modes: speech-to-text, streaming transcript deltas
Released: 28 Jul 2026

OpenAI published no system card for GPT Live Transcribe. Released 28 July 2026 alongside GPT Transcribe, it is OpenAI's recommended model for transcribing live audio from a microphone, call or media stream. It streams transcript deltas as speech arrives and a final transcript for each committed turn, with a delay setting that trades latency for accuracy, plus free-form context, keyword hints and several expected languages. model page; changelog; realtime transcription guide

gpt-live-transcribe

Typical price: $0.017 / min
Weights: Closed
Released: 28 Jul 2026

speech-to-text
streaming transcript deltas
tunable latency (delay setting)
context prompt and keyword hints
multiple language hints / code-switching