ChatGPT Audio Uploads: Transcripts, Cost, and Catches

That recording you’ve been meaning to listen to again? As of October 6, ChatGPT can take an existing audio file and help you work through it. Attach the file, ask for a transcript or summary, then ask questions about what was said. This is an audio-upload feature OpenAI announced on October 6, not a new way to record a meeting or have a live voice conversation.

The practical use is finding the part you need without replaying an entire lecture, interview, or meeting. You could attach a recording and ask: “Transcribe this, then list the decisions made and any questions nobody answered.” Treat the result as a way to locate the important bits, not as an official record of who agreed to what. OpenAI warns that transcripts can contain errors and speaker identification may be unreliable. Check names, numbers, and consequential statements against the audio.

There’s a price-of-entry catch: OpenAI’s upload guide says audio uploads are for paid subscriptions and workspaces, including Enterprise, not the Free plan. Plus costs $20 a month, but access can depend on your region, workspace settings, app version, and selected model. Check whether the option appears in your account before paying just for this. Supported audio formats include MP3, WAV, and M4A; the limit is 512 MB per file, and very long recordings may time out. A video file doesn’t count just because it has sound.

One more thing before uploading somebody else’s conversation: get appropriate permission. OpenAI says files saved in Library can remain there separately from a chat, so deleting the conversation may not delete its saved recording. On eligible individual accounts, files and chats may also be used to improve models if “Improve the model for everyone” is on; business-workspace defaults differ.

For a recording you already have permission to use, this could save a lot of replaying. Just don’t let a neat-looking summary persuade you it heard every word correctly.