Every recorder saw an identical scene for exactly one minute, at its own default settings, on the same Mac. Then each file was read back with ffprobe and each run had its CPU and memory sampled once a second. No settings were tuned, no favourites were picked. The raw data is downloadable at the bottom.
The smallest file was 14.7 MB per minute (ffmpeg (AVFoundation, libx264 defaults)); the largest was 222.8 MB per minute (macOS built-in (QuickTime / ⌘⇧5)). Resolution, frame rate and bitrate explain most of the gap, and they are all in the table.
| Recorder | MB / min | Bitrate | Output | fps | Audio | CPU added | Peak RAM | Stop → file |
|---|---|---|---|---|---|---|---|---|
| ffmpeg (AVFoundation, libx264 defaults) reference: -f avfoundation, 30 fps, libx264 defaults, no audio | 14.7 | 2.0 Mbps | 2704×1756 · H264 High | 30 | none | +6.7 pts | 1,776 MB | 1 s |
| OwnClipours default (full screen, mic + system audio as separate stems) | 57 | 7.4 Mbps | 2704×1756 · H264 High | 29 | AAC | +10.4 pts | 252 MB | 2 s |
| macOS built-in (QuickTime / ⌘⇧5) default, no audio | 222.8 | 29.7 Mbps | 2704×1756 · H264 Main | 57 | none | +8.7 pts | 46 MB | 1 s |
CPU added = average whole-machine CPU during the recording minus the 25.4% the scene alone used with nothing recording. Stop → file = seconds from pressing stop until a playable file existed; recorders that open an editor first are measured to the end of their default export.
Not a product, a reference point. ffmpeg capturing the display through Apple’s AVFoundation with its stock software encoder shows what a plain, unoptimised software pipeline costs on the same scene.
Recorded with the settings a fresh install uses: full display, microphone and system audio captured as separate tracks, and the take opening straight into the editor when you stop. That is why the file is larger than the built-in recorder’s: it keeps a higher bitrate so text stays sharp when you zoom in later, and it carries two audio tracks.
The CPU figure includes the live audio clean-up and the on-device work that runs while recording. It is the most expensive recorder in this table to run and the only one whose output needs no second application afterwards.
This is the recorder every Mac ships with: the ⌘⇧5 toolbar and QuickTime share the same capture pipeline, so one run covers both. It writes H.264 in a .mov container, records no audio unless you pick a microphone first, and leans hard towards quality over size: it keeps the display’s full frame rate and a very high bitrate, which is why its files are by far the largest here. That is the file most people end up compressing or uploading for minutes.
It costs almost nothing to run because the encoding is handled by the system rather than the app. What you give up is everything after the recording: there is no editor, no cursor or zoom effects and no way to fix a take without another app.
Machine. MacBook Pro, Apple M4 Max, macOS 26.6.2, 3024×1964 built-in display. Nothing else running except the browser showing the scene.
Scene. identical 60 s browser scene: static UI, scrolling text, drifting gradient, six moving shapes, running clock. It is a single web page shown full-screen, and it is public: open the benchmark scene and record it yourself.
Settings. Each recorder exactly as installed, recording the full display for 60 seconds, with audio left at its default. The table shows which files carry an audio track. Where a recorder needs an export step before a file exists, its default export was used.
Reading the file. Resolution, frame rate, codec and bitrate come from ffprobe. MB per minute is file size divided by measured duration.
CPU and memory. Sampled once a second from the operating system for the whole machine and for the recorder’s own processes, then averaged over the minute. The baseline run had the scene open and no recorder.
What this does not measure. Editing features, sharing, cursor effects, or how the recordings look side by side. It measures what the recorders write to disk and what they cost to run.
Because defaults are what most people record with. Every recorder here can be tuned; the question this page answers is what you get when you press record for the first time.
A screen recording is usually uploaded, shared or archived. At 30 MB per minute a ten-minute recording is 300 MB; at 10 MB per minute it is 100 MB. Over a year of recordings that difference is disk space, upload time and, on cloud plans, money.
No. Smaller files mean stronger compression, and stronger compression can blur text and smear motion. That is why the table lists resolution, frame rate and bitrate next to the size, and why the same scene is used for every recorder so the outputs can be compared frame for frame.
One sample per second from the operating system during the 60-second recording, for the whole machine and for the recorder’s own processes. The scene itself costs CPU, so the table reports what the recorder added on top of a baseline run with the scene open and nothing recording.
Yes. The scene is a single HTML page linked on this page, and the method is described step by step. Record it for 60 seconds with any recorder at its default settings, then read the file with ffprobe.
Every number on this page, as JSON, licensed CC BY 4.0. Cite it, chart it, or run the scene through your own recorder and add a row.
/benchmarks/data.json →