OpenAI Whisper is a speech-recognition model that transcribes and translates audio. Unlike a hosted transcription app, Whisper is published as open-source code and model weights, so you can run it on your own hardware, embed it in an application, or call OpenAI's hosted whisper-1 endpoint through the API. That flexibility is why it appears inside many transcription products even when the brand on the screen is different.
The model is multilingual, handles noisy audio and can translate speech into English as well as transcribe it. The hosted API accepts audio files up to 25MB and bills by the minute at $0.006, with no subscription. Running it locally removes per-minute fees but shifts the cost to hardware and engineering: you need to install the runtime, choose a model size, manage audio conversion and handle long files by splitting them.
Whisper is not a finished consumer app. There is no meeting bot, speaker-labelled dashboard or team workspace. It is a building block for developers and a free offline option for privacy-sensitive work. OpenAI has also scheduled whisper-1 for shutdown on February 26, 2027, recommending gpt-live-transcribe or gpt-transcribe instead, so new integrations should plan for migration rather than treating whisper-1 as a permanent endpoint.
