OpenAI: Whisper Large V3

OpenAI: Whisper Large V3 is an automatic speech recognition model from OpenAI, distributed as open source. It sits in the transcription category and is the large version of the Whisper family, which means it is the variant OpenAI positions for the widest language coverage and the most demanding audio. If you already route other OpenAI models through api14, this one comes from the same provider and follows the same conventions.

The model takes audio as its only input and returns a transcription as its output. According to OpenAI, it handles both transcription and translation, so you can either write down speech in its original language or render it into another one. It covers more than 99 languages and accepts common audio formats, including mp3, mp4, wav and webm. There is no context window figure to plan around here, and the maximum output is listed as zero, so treat the audio file itself, not a token budget, as the thing you size your requests against.

Those capabilities point at a fairly specific set of jobs. Meeting and call recordings can be converted into text for search, review or archival. Interviews and user research sessions can be transcribed in bulk, including material in languages your team does not speak, since the model covers 99+ languages. Podcasts, lectures and video files in mp3, mp4, wav or webm can be turned into written records or subtitles. Because translation is part of the same model, you can send non-English audio and get output in a target language without running a separate translation step.

You call it through api14's OpenAI-compatible API using the model id "openai/whisper-large-v3". Because the interface matches OpenAI's, existing OpenAI client libraries and request shapes work with only the base URL and key changed. The model's supported parameter list is empty, so do not expect to tune sampling, temperature or similar knobs; send the audio and read back the transcription. Check the live price table on this page for current rates before you commit to a workload.

One practical caveat: this is a transcription model, not a chat or reasoning model. It does not take text prompts as input and does not produce conversational output, so tasks like summarising a transcript, answering questions about it or generating code belong to a text model. A common pattern is to chain the two, using Whisper Large V3 to produce the transcript and a separate language model to work on the resulting text. If your input is already text, or your audio is in a language outside the supported set, this model is not the right fit.