What is Whisper
Whisper is OpenAI’s general-purpose multilingual speech-recognition model. The open-source model and inference code are released under the MIT License and support multilingual transcription, speech translation into English, and language identification.
Pricing
As of July 28, 2026, downloading and self-hosting the open-source Whisper code has no OpenAI licence fee, although users pay their own compute and operations costs. OpenAI’s hosted whisper-1 API costs US$0.006 per audio minute through the transcription and translation endpoints.
Hosted API features and limits
The hosted model accepts audio uploads up to 25 MB and provides transcription or translation rather than native realtime streaming. It remains useful where teams want the established Whisper interface without running their own model infrastructure.
OpenAI also publishes newer realtime speech models separately; their distinct pricing should not be confused with the whisper-1 rate.
Official sources: OpenAI Whisper repository, whisper-1 model and pricing, Whisper API FAQ.