YouTube stereo audio at 384 kbps in a 3-minute Short is 8.64 MB

YouTube's recommended upload settings (read 2026-10-09) list stereo audio at 384 kbps, mono at 128 kbps and 5.1 at 512 kbps, with a 48 kHz sample rate. For a 180-second Short that is 384 x 180 / 8 = 8,640 KB (8.64 MB) of stereo audio, 2.88 MB mono and 11.52 MB for 5.1. Next to video at the 1080p rate of 8 Mbps (180 MB for 180 seconds) the audio is about 4.6% of the file.

This is arithmetic on YouTube's recommendations, not a Sume measurement. Sume does not expose an audio bitrate: the trim API keeps audio as AAC on exact precision without a bitrate field. What Sume does let you pick is the sample rate and channel layout when you pull the audio out of a video with audio detach.

The numbers

Size in decimal megabytes is kbps x seconds / 8 / 1,000.

What Sume writes

Audio detach takes one media.sume.com video and returns a new audio artifact. The default is sample-exact wav (pcm_s16le); the other format is mp3 at 128 kbps. Channels are source or mono, and sample_rate is 16000, 44100 or 48000, or inherited from the source if omitted. The price is $0.01 per job.

A stereo 48 kHz 16-bit wav is 48,000 x 16 x 2 = 1,536 kbps, four times the YouTube number, so 180 seconds is 34.56 MB. That file is for editing and lip-sync, not for upload. An mp3 at 128 kbps stereo is 2.88 MB for 180 seconds, which is below the 384 kbps YouTube recommends. Neither is the file YouTube sees; the platform receives the audio inside your final MP4.

Where the audio in the final MP4 comes from

With video trim exact, kept audio is remuxed as AAC. A Timeline render builds the output from one audio spine (the wav you supply or join) and the video slots, and the render docs do not state an audio bitrate. So the only claim this post makes is the one the YouTube page makes: 48 kHz, stereo, and 384 kbps if you can choose. Check the delivered file with your own tool before upload if the number matters to you.

The same page says the audio codec should be AAC-LC, Opus or Eclipsa Audio, and that the channel layout is stereo or stereo plus 5.1. Sume's trim doc names AAC on exact precision; it does not name Opus or Eclipsa.

A short checklist

• Pick 48000 for sample_rate when you detach audio you plan to re-join
• Join with timeline audio concat (wav, $0.01) and render once; avoid re-encoding mp3 twice
• Mono sources are fine: 128 kbps mono is the YouTube line for them
• Detach channels is source or mono, so there is no stereo downmix option