← All models

GPT-4o Transcribe

Published Publisher OpenAIgpt-4o-transcribe

Model Overview

GPT-4o Transcribe is a speech-to-text model based on GPT-4o that transcribes audio into text through the OpenAI API. OpenAI released the model on March 20, 2025. (openai.com)

Capabilities and Uses

  • Supports multilingual speech recognition. OpenAI states that, compared with the original Whisper model, it improves language identification and accuracy, with lower word error rates on multiple benchmarks; it also handles accents, background noise, and variations in speaking speed better. (openai.com)
  • Can be used for transcribing recordings, generating text records, and transcription in speech applications; existing integrations support streaming output for file transcription and real-time transcription. (developers.openai.com)

Limitations and Status

  • Transcription results can still contain errors, and important content should be checked by a human. The model does not support image or video input and does not generate audio; its context window is 16,000 tokens, and its maximum output is 2,000 tokens. (developers.openai.com)
  • The file transcription endpoint has a file size limit of 25 MB; larger recordings need to be compressed or split. (developers.openai.com)
  • As of November 19, 2026, OpenAI has marked the model as deprecated and plans to remove it from the API on February 26, 2027; OpenAI recommends migrating to gpt-live-transcribe or gpt-transcribe depending on the use case. (developers.openai.com)

Pricing

Date checked: November 19, 2026. The official OpenAI API pricing page lists prices in US dollars (USD): input at $2.50 per million tokens, output at $10.00 per million tokens; the estimated transcription cost is $0.006 per minute. This is an estimate, not a separately guaranteed fixed per-minute billing rate. (developers.openai.com)

Official materials differ in how they label input modalities: the pricing table does not break down modalities in this model's input column; OpenAI's Realtime transcription example lists audio input at $6.00 per million audio tokens, text input at $2.50 per million text tokens, and text output at $10.00 per million text tokens. The applicable audio input rate should be verified against account billing and the latest OpenAI guidance. (developers.openai.com)

Official Sources