Whisper large-v3
OpenAI · USA · 2023
The open-source ear of the AI era — speech recognition in 100 languages, everywhere.
Whisper is OpenAI's open-source speech-recognition model, robust to accents, noise and technical vocabulary across roughly 100 languages. Released under MIT license, it became the default transcription engine of the internet — embedded in meeting tools, subtitling pipelines, voice assistants and countless robots' listening stacks. Its profile in the catalogue covers two different things that are easy to confuse. One is the weights: openai/whisper-large-v3, MIT-licensed, downloadable by anyone, impossible to withdraw from those who already have them. The other is whisper-1, the hosted endpoint OpenAI sells in its own API at $0.006 per minute. On 26 August 2026 OpenAI deprecated that endpoint and set its removal for 26 February 2027, together with the three GPT-4o transcription models; the recommended replacements are gpt-transcribe and gpt-live-transcribe. What makes the shutdown awkward is that OpenAI's own transcription guide, as of 3 September 2026, still routes three capabilities to whisper-1: word-level timestamps, srt and vtt subtitle output, and translation of a recording into English. The recommended replacements do not offer them. Anyone generating subtitles through OpenAI is therefore pointed at a model with a shutdown date and no like-for-like successor inside the platform — while the same work can be done indefinitely on the open weights, on their own hardware, for the cost of the electricity.
▸News
▸Videos
No videos yet.
▸Reviews
No reviews yet. Be the first!