Which screen recorder exports fastest?

Same clip, same edit, same machine. We time each app's export, then divide by what plain ffmpeg needs on that machine — so 1.35× means 35% longer than a raw re-encode, on any hardware.

33Submissions

71Runs

5Tools timed

12Builds

4Machines

3Platforms

Data as of 2026-10-04 · every export re-checked in pixels and audio before it counts

01 — Ranking

Ratio to ffmpeg

Platform

GPU

All 71 runs — 5 tools across 12 builds, 4 machines, 3 platforms. Every machine combined.

↓ Lower is better — a shorter bar is a faster export

ffmpeg 1.00×9×7×5×3×

openscreen

2.0.0

1.35×

screen-studio

4.0.1-4897

2.00×

focusee

2.4.1 (macOS) · 2.3.5 (Windows)

4.27×

cap

0.6.0

4.79×

recordly

1.4.0 (macOS, Windows) · 1.3.3 (Linux)

7.95×

Bars start where ffmpeg finishes: their length is what the tool spends on top of a raw re-encode. The figure beside each is the full ratio, floor included.

Read the ratio, not the seconds. 2× means the export took twice as long as ffmpeg did on that machine — not twice as fast.

ffmpeg is the 1×. It re-encodes the clip and composites nothing, so a tool that paints a wallpaper, rounds the corners and animates a zoom sits above it.

No tool is the reference. Every export is divided by an ffmpeg transcode on its own machine, so no submission has to contain any particular tool.

Cross-check: where a build was measured on more than one machine, its cost moves by 239.35% at the median — widest: focusee, 1.15× on Ryzen 7 5800X and 10.73× on Ryzen 5 7520U.

5 older builds not shown: each row is the newest build measured, and the ones it replaced keep their own figures. Every build measured →

02 — Spread across machines

macOSLinuxWindows

One dot per run, placed at its ratio; colour is the platform. Hover a dot for the machine and the seconds.

1× = ffmpeg itself

focusee

openscreen

cap

screen-studio

recordly

1×2×3×4×5×6×7×8×9×10×11×12×13×14×15×16×17×

Export time ÷ the same machine's ffmpeg transcode. 2× means twice as long, not twice as fast.

03 — Every run

71 runs

Every run, with the machine that produced it. Cost compares across rows; seconds and frame rate don't — they describe one machine.

Machine Platform Tool

Tool	Machine	OS	Cost	sw/hw floor	Export s	fps	CPU·s	Cores	RSS MiB	Bg load	Fidelity	Driven
openscreen 2.0.0	Apple M1 · 16 GB	darwin	1.074×	1.34×	20.19 ±0.04	178.3	30.39	1.51	858	162%	full	cli
screen-studio 4.0.1-4897	Apple M1 · 16 GB	darwin	1.937×	1.33×	36.19 ±0.07	99.5	56.16	1.55	1990	78%	full	cdp+menu
focusee 2.4.1	Apple M1 · 8 GB	darwin	0.932×	1.35×	16.95 ±0.01	212.4	35.36	2.09	439	112%	full	ax+project
openscreen 1.11.0-rc.1	Apple M1 · 8 GB	darwin	1.037×	—	18.91 ±0.01	190.3	27.49	1.45	785	14%	full	cli
openscreen 1.11.0-rc.1	Apple M1 · 8 GB	darwin	1.037×	1.25×	18.90 ±0.03	190.5	27.81	1.47	785	49%	full	cli
openscreen 1.11.0-rc.1	Apple M1 · 8 GB	darwin	1.038×	—	18.90 ±0.01	190.5	27.71	1.47	786	50%	full	cli
openscreen 1.11.0-rc.1	Apple M1 · 8 GB	darwin	1.039×	1.30×	18.91 ±0.01	190.4	28.30	1.50	739	96%	full	cli
openscreen 1.11.0-rc.1	Apple M1 · 8 GB	darwin	1.043×	—	19.03 ±0.03	189.2	27.56	1.45	784	62%	full	cli
openscreen 2.0.0	Apple M1 · 8 GB	darwin	1.065×	—	19.41 ±0.01	185.4	27.07	1.39	809	12%	full	cli
openscreen 2.0.0-rc.12	Apple M1 · 8 GB	darwin	1.111×	—	20.20 ±0.05	178.2	28.69	1.42	854	37%	full	cli
openscreen 1.10.0	Apple M1 · 8 GB	darwin	2.023×	—	36.80 ±0.01	97.8	8.54	0.23	719	81%	full	cli
screen-studio 4.0.1-4897	Apple M1 · 8 GB	darwin	2.045×	—	37.45 ±0.06	96.1	56.78	1.52	1075	54%	full	cdp+menu
screen-studio 4.0.1-4897	Apple M1 · 8 GB	darwin	2.066×	—	37.82 ±0.07	95.2	56.37	1.49	1260	63%	full	cdp+menu
screen-studio 3.7.5-4595	Apple M1 · 8 GB	darwin	4.681×	1.23×	85.73	42.0	219.99	2.57	1450	76%	full	cdp+menu
cap 0.6.0	Apple M1 · 8 GB	darwin	1.186×	1.21×	21.60 ±0.03	166.7	15.29	0.71	1023	35%	full	cli
cap 0.6.0	Apple M1 · 8 GB	darwin	1.192×	—	21.69 ±0.04	166.0	15.35	0.71	946	27%	full	cli
cap 0.5.9	Apple M1 · 8 GB	darwin	1.210×	—	22.01 ±0.03	163.6	24.39	1.11	1485	58%	full	cli
cap 0.5.9	Apple M1 · 8 GB	darwin	1.224×	—	22.30 ±0.18	161.4	24.58	1.10	1487	31%	full	cli
cap 0.5.9	Apple M1 · 8 GB	darwin	1.236×	—	22.52 ±0.02	159.9	24.69	1.10	1552	41%	full	cli
cap 0.5.9	Apple M1 · 8 GB	darwin	1.255×	—	22.88 ±0.01	157.3	26.10	1.14	1588	51%	full	cli
recordly 1.3.3	Apple M1 · 8 GB	darwin	2.626×	—	47.74 ±0.01	75.4	90.61	1.90	1232	93%	full	cdp
recordly 1.4.0	Apple M1 · 8 GB	darwin	2.690×	—	49.30 ±0.02	73.0	65.24	1.32	1190	63%	full	cdp
focusee 2.3.5	Ryzen 5 7520U	win32	10.728×	2.12×	207.79 ±3.29	17.3	467.23	2.25	3340	162%	full	uia
openscreen 1.11.0-rc.1	Ryzen 5 7520U	linux	1.124×	1.55×	27.27 ±0.02	132.0	69.00	2.53	968	47%	full	cli
openscreen 2.0.0	Ryzen 5 7520U	linux	1.423×	1.83×	26.97 ±0.04	133.5	60.00	2.22	1020	5%	full	cli
openscreen 1.11.0-rc.1	Ryzen 5 7520U	win32	1.441×	2.10×	30.73 ±0.06	117.2	9.61	0.31	896	64%	full	cli
openscreen 1.11.0-rc.1	Ryzen 5 7520U	win32	1.457×	2.08×	30.63 ±0.12	117.5	9.28	0.30	873	49%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	win32	1.582×	—	32.76 ±0.18	109.9	10.94	0.33	1004	77%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	win32	1.671×	2.16×	35.74 ±0.35	100.7	11.09	0.31	939	75%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	win32	1.794×	2.20×	34.52 ±0.74	104.3	11.08	0.32	966	91%	full	cli
openscreen 1.11.0-rc.1	Ryzen 5 7520U	win32	1.839×	2.19×	34.93 ±0.20	103.1	10.02	0.29	877	60%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	win32	1.907×	—	39.03 ±0.78	92.2	10.50	0.27	984	50%	full	cli
openscreen 2.0.0	Ryzen 5 7520U	win32	1.929×	2.12×	35.44 ±2.35	101.6	9.91	0.28	919	38%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	win32	2.015×	2.12×	38.55 ±0.09	93.4	10.64	0.28	947	64%	full	cli
openscreen 2.0.0	Ryzen 5 7520U	win32	2.130×	2.29×	42.88 ±0.44	84.0	16.46	0.38	1586	118%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	linux	3.300×	1.60×	76.84 ±0.01	46.8	119.00	1.55	996	138%	full	cli
openscreen 1.10.0	Ryzen 5 7520U	linux	4.085×	2.03×	77.85 ±0.15	46.2	121.00	1.55	993	51%	full	cli
cap 0.5.9	Ryzen 5 7520U	linux	2.416×	1.59×	57.02	63.1	166.00	2.91	1232	47%	full	cli
cap 0.5.9	Ryzen 5 7520U	linux	2.500×	1.62×	57.31 ±0.03	62.8	165.00	2.88	1231	142%	full	cli
cap 0.6.0	Ryzen 5 7520U	linux	2.734×	1.86×	51.68 ±0.03	69.7	158.00	3.06	764	14%	full	cli
cap 0.6.0	Ryzen 5 7520U	linux	2.861×	2.02×	54.71 ±0.26	65.8	171.00	3.13	704	83%	full	cli
cap 0.5.9	Ryzen 5 7520U	linux	2.991×	1.95×	57.14 ±0.15	63.0	166.00	2.91	1231	54%	full	cli
cap 0.6.0	Ryzen 5 7520U	win32	6.727×	1.93×	143.53 ±2.25	25.1	51.88	0.36	1378	101%	full	cli
cap 0.6.0	Ryzen 5 7520U	win32	6.883×	2.05×	135.27 ±0.91	26.6	51.81	0.38	1384	43%	full	cli
cap 0.6.0	Ryzen 5 7520U	win32	6.994×	2.14×	130.26 ±0.08	27.6	48.05	0.37	1378	50%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	7.694×	—	149.85 ±0.32	24.0	102.17	0.68	2395	85%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	8.251×	2.14×	172.42 ±2.28	20.9	103.05	0.60	2351	60%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	8.293×	—	175.28 ±0.86	20.5	104.52	0.60	2377	76%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	8.723×	2.11×	185.29 ±0.60	19.4	103.88	0.56	2354	71%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	8.778×	1.99×	185.62 ±1.79	19.4	103.58	0.56	2348	75%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	8.845×	2.13×	190.94 ±0.56	18.9	105.34	0.55	2358	67%	full	cli
cap 0.5.9	Ryzen 5 7520U	win32	9.345×	2.27×	181.18 ±3.22	19.9	104.44	0.58	2359	72%	full	cli
recordly 1.4.0	Ryzen 5 7520U	win32	6.131×	1.08×	226.44 ±0.39	15.9	601.78	2.66	2369	57%	full	cdp
recordly 1.4.0	Ryzen 5 7520U	win32	6.893×	1.28×	216.65 ±7.29	16.6	613.65	2.83	2324	48%	full	cdp
recordly 1.3.3	Ryzen 5 7520U	win32	10.446×	—	197.62 ±5.50	18.2	534.85	2.71	2678	33%	full	cdp
recordly 1.3.3	Ryzen 5 7520U	win32	10.654×	—	203.15 ±2.10	17.7	560.14	2.76	2662	45%	full	cdp
recordly 1.3.3	Ryzen 5 7520U	linux	13.710×	1.58×	323.90 ±2.28	11.1	505.00	1.56	2403	138%	full	cdp
recordly 1.3.3	Ryzen 5 7520U	linux	16.997×	1.96×	324.46 ±3.96	11.1	505.00	1.56	2459	39%	full	cdp
focusee 2.3.5	Ryzen 7 5800X	win32	1.145×	2.23×	8.67 ±0.02	415.4	46.84	5.40	2195	326%	full	uia
focusee 2.3.5	Ryzen 7 5800X	win32	1.157×	2.01×	8.63 ±0.05	417.0	42.45	4.92	2215	177%	full	uia
openscreen 2.0.0	Ryzen 7 5800X	win32	1.195×	1.96×	9.05 ±0.01	397.6	4.52	0.50	796	192%	full	cli
openscreen 1.10.0	Ryzen 7 5800X	win32	1.204×	—	9.38	383.6	3.88	0.41	774	69%	full	cli
openscreen 1.11.0-rc.1	Ryzen 7 5800X	win32	1.242×	2.01×	9.37 ±0.02	384.2	4.17	0.44	770	87%	full	cli
openscreen 1.11.0-rc.1	Ryzen 7 5800X	win32	1.262×	2.09×	9.75 ±0.30	369.2	4.80	0.49	773	272%	full	cli
openscreen 1.10.0	Ryzen 7 5800X	win32	1.292×	—	10.09 ±0.01	356.7	4.50	0.45	804	69%	full	cli
cap 0.5.9	Ryzen 7 5800X	win32	2.937×	—	22.97 ±0.49	156.7	41.84	1.82	1379	96%	full	cli
cap 0.5.9	Ryzen 7 5800X	win32	3.145×	—	24.48 ±0.01	147.1	44.20	1.81	1387	128%	full	cli
cap 0.6.0	Ryzen 7 5800X	win32	8.261×	2.03×	62.38 ±3.88	57.7	27.33	0.44	1106	81%	full	cli
recordly 1.3.3	Ryzen 7 5800X	win32	3.400×	—	26.39 ±0.09	136.4	82.42	3.12	1606	86%	full	cdp
recordly 1.3.3	Ryzen 7 5800X	win32	3.702×	—	28.79 ±0.01	125.1	147.68	5.13	1524	48%	full	cdp
recordly 1.4.0	Ryzen 7 5800X	win32	7.185×	1.94×	54.84 ±0.08	65.6	162.72	2.97	1695	197%	full	cdp

Export carries ± the median absolute deviation of that machine's own scoring runs: the spread its own repeats landed in.

Cores is CPU-seconds ÷ export time: how many cores the export kept busy. No GPU time is measured, on either side of the ratio.

Bg load is everything running that was not the tool under test, with the tool's own processes subtracted. Where two rows on one machine differ sharply, their seconds are not comparable — and Cost does not repair it: the floor runs on the fixed-function encoder block, which barely throttles, while the compositing under test runs on cores and shaders, which do. How far that goes, measured.

04 — Per build

Every build measured, not only the newest — cost, coverage and the range each has been seen inside. The ranking above shows one row per tool; this is the rest.

Tool	Build	Cost	Runs	Platforms	Range	Versions
cap	0.6.0	4.794×	8	macOS, Linux, Windows	1.186–8.261×	0.6.0
cap	0.5.9	3.858×	16	macOS, Linux, Windows	1.21–9.345×	0.5.9
focusee	2.4.1	0.932×	1	macOS	0.932–0.932×	2.4.1
focusee	2.3.5	5.94×	3	Windows	1.145–10.728×	2.3.5
openscreen	2.0.0	1.354×	6	macOS, Linux, Windows	1.065–2.13×	2.0.0
openscreen	2.0.0-rc rc	1.111×	1	macOS	1.111–1.111×	2.0.0-rc.12
openscreen	1.11.0-rc rc	1.221×	11	macOS, Linux, Windows	1.037–1.839×	1.11.0-rc.1
openscreen	1.10.0	2.163×	10	macOS, Linux, Windows	1.204–4.085×	1.10.0
recordly	1.4.0	5.484×	4	macOS, Windows	2.69–7.185×	1.4.0
recordly	1.3.3	8.031×	7	macOS, Linux, Windows	2.626–16.997×	1.3.3
screen-studio	4.0.1-4897	1.997×	3	macOS	1.937–2.066×	4.0.1-4897
screen-studio	3.7.5-4595	4.681×	1	macOS	4.681–4.681×	3.7.5-4595

05 — Roster

Membership follows what a tool is built for, not whether an adapter exists. A cell that is not a member says who says so.

Tool	Windows	Linux
OpenScreen
Screen Studio	n/a	n/a
Cap
Recordly		degraded
FocuSee		n/a

Recordly on Linux — per the vendor's own documentation: the rendered cursor is lost, so S3 is n/a here

06 — Method

Same clip, same edit

Every tool is handed the same recording and asked for the same demo — wallpaper, rounded corners, three zooms, a drawn cursor, a webcam inset — out at 1080p60 H.264.

Only the export is timed

The clock starts when the export is launched and stops when the file is finished. Opening the app, loading the project and setting the presets happen before it.

Divided by ffmpeg, checked in the pixels

The same machine re-encodes the same clip with plain ffmpeg, minutes apart: that run is the 1×. Each export is then re-opened to confirm the size, the length, the effects and the sound are really there.

Read the protocol in full →

07 — Submit a run

Every new machine adds a hardware combination nobody has measured yet. Yours takes about twenty minutes.

node bench.mjs run --bundle commons-upload

01Clone the repo and install the adapters for the tools you own.
02Run the bench on an idle machine. It measures its own ffmpeg floor beside every leg.
03Open a pull request with the produced JSON in submissions/. Validation runs on the PR.