whisper-1 is OpenAI's speech-to-text model, corresponding to Whisper large-v2 when the official API was released. It is suitable for organizing speech from interviews, courses, voice messages, and audiovisual materials into original-language text, with text, JSON, or subtitle formats selected according to the workflow. On this platform, you can submit files through the audio transcription entry point to integrate recording processing into content production and document organization workflows.
Clarify capacity, input/output, and invocation methods before selecting a model.
Native format details and invocation parameters are listed separately; this entry point focuses on file transcription and is not equivalent to real-time voice conversation.
Learn what whisper-1 can bring to your work.
whisper-1's transcription task converts spoken audio into original-language text rather than translating it into English by default. This is suitable for interviews, courses, and dictated materials that need to retain the original wording. The generated text can serve as a basis for subsequent search, excerpting, and editing, while summarization and rewriting should be handled as separate processing steps.
The same transcription workflow can select output formats according to the use case: text is suitable for direct entry into an editor, JSON makes text easy for programs to read, and SRT and VTT are suitable for subtitle production. Format selection should be designed around the final deliverable, avoiding saving plain text first and then additionally rebuilding subtitle structures and processing workflows.
Calls are centered on audio files, so recordings do not need to be converted into chat messages. Requests can include language and prompt text to help applications organize transcription context; prompts should be prepared around the recording content rather than used as free-form creative instructions. This makes it easier to connect uploading, transcription, storage, and human proofreading into a stable workflow.
Start with specific tasks to find where the model can be useful.
Submit interview recordings or oral materials as files, choose text or JSON output, and obtain transcripts that are easy to search and edit. Editors can use them to locate topics, organize quotations, and listen back to verify names, organization names, and key original statements; article structure and viewpoint extraction can be handled separately after transcription is complete.
Choose SRT or VTT output for course recordings or video materials, and send the results into subtitle editing or playback workflows. Before delivery, check sentence breaks and the correspondence between text and timing against the original material; the model handles speech transcription, while subtitle layout, visual styling, and final video export are still completed by production tools.
Convert user voice messages into text for storage in ticketing systems or knowledge bases. Applications can read the text in JSON and associate the transcription with the original audio, making it easier for staff to review and listen back. Classification, priority assessment, and reply generation should be handled in separate processing steps rather than treating transcription results directly as business decisions.
Choose based on task complexity, input materials, and expected results.
If the core requirement is to process existing recordings and deliver original-language text or subtitles, whisper-1 has clear task boundaries. It is not a general question-answering model, nor is it a speech synthesis model: it will not read text aloud as audio, and it should not be expected to handle all the work of automatically writing meeting notes. Complete transcription first, then follow with summarization or editing workflows to make it easier to verify the original wording.
whisper-1 is the API invocation name, while large-v2 is the corresponding native version at the time of official release; it should not be understood as a general term for all newer Whisper versions. When comparing other transcription models, use the same recordings to check proper nouns, sentence breaks, and required output formats; if the goal is English translation, also distinguish between transcription in the original language and independent translation tasks.
From a small-scale task to formal integration.
Define the objective, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand the output quality and scope of capabilities.
Answers to common questions about using whisper-1.
whisper-1 is the model name used when calling the API, and OpenAI's Whisper API release notes map it to large-v2. They describe the invocation name and native version respectively; whisper-1 should not be treated as a generic alias for all Whisper versions, nor should this be taken to mean that it is the latest version.
The task for the audio transcription endpoint is to output text in the original language, not to translate it into English by default. Officially, transcription in the original language and translation into English are treated as separate tasks. Therefore, Chinese recordings should be processed using the Chinese transcription workflow; if an English transcript is needed, a separate translation step should be arranged.
The call parameters provide srt and vtt output options for subtitle workflows; for plain-text tasks, text or json can be selected. After subtitles are generated, the content, line breaks, and timing should still be checked against the audio before handing them off to subtitle editing or playback tools; they should not be regarded as a finished product directly.
prompt is hint text in a transcription request and should not be treated as a general-purpose writing instruction. The core deliverable of whisper-1 remains an audio transcript. When an article, summary, or action items are needed, it is recommended to retain the original transcription result and then pass it to a separate text-processing step, avoiding confusion between the original words and processed content.
When calling /v1/audio/transcriptions directly, you need to submit a binary file and use whisper-1. If using an MCP client, you can choose its tool for transcribing audio from a URL. The inputs are organized differently for the two methods; do not submit a URL string directly as the binary file field in HTTP.
Model information · Updated: 2026-10-01. See the API and pricing sections for call parameters and billing rules.
Start with a clear goal and judge from real results whether it suits your work.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs