GPT-4o Transcribe

GPT-4o Transcribe, released on 20 March 2025, was OpenAI's first attempt to replace Whisper with a transcription model built on its flagship multimodal line. Its selling point was accuracy: a lower word error rate and better language recognition than the original Whisper models, on the same single endpoint developers already used. It takes audio and text in and returns text only, with a 16,000-token context window and up to 2,000 output tokens; its knowledge cutoff is 1 June 2024. Billing can be read two ways, which is worth knowing before comparing it with anything else. The per-token rate is $2.50 per million audio input tokens and $10 per million text output tokens; the same usage expressed by recording length is $0.006 per minute — exactly what OpenAI charges for whisper-1, the model it was meant to replace. Since 28 July 2026 the recommended entry point is GPT Transcribe at $0.0045 per minute, and the cheaper sibling of this model, GPT-4o mini Transcribe, costs $0.003 per minute. GPT-4o Transcribe therefore sits in an awkward middle: it is neither the accurate default nor the cheap option. One detail sets it apart from every other transcription model OpenAI sells, and not in its favour. Its snapshot list contains a single entry — the bare alias `gpt-4o-transcribe`, with no date attached. The cheaper mini version has two dated snapshots and lets a customer pin one; the paid-double flagship of the pair does not. Anyone who needs a frozen, auditable version of the model they are billed for cannot get one here. As of August 2026 OpenAI has announced no shutdown date for the model, and it remains available through the audio transcriptions endpoint and inside Realtime sessions.

▸Videos

No videos yet.

▸Reviews