benchmarks.md

Linux

_{70b68ca083f6 · p50 latency, lower is faster. Full Raw Benchmarks}

Scenario	rmux	tmux	zellij	vs tmux
Create background session (default shell)	3.565 ms	7.565 ms	-	🟢 2.1x faster
List sessions	1.744 ms	3.374 ms	16.358 ms	🟢 1.9x faster
Split pane left/right	3.456 ms	3.386 ms	63.832 ms	≈ same speed
Split pane top/bottom	3.418 ms	3.399 ms	63.769 ms	≈ same speed
Send keys to pane/window	1.332 ms	3.422 ms	31.715 ms	🟢 2.6x faster
Capture visible output	1.305 ms	3.476 ms	16.806 ms	🟢 2.7x faster
Capture full scrollback	3.470 ms	3.514 ms	-	≈ same speed
List tabs/windows	3.646 ms	3.939 ms	15.966 ms	≈ same speed
Grow pane by 1 cell	1.715 ms	3.461 ms	-	🟢 2.0x faster
Terminate session	2.060 ms	3.392 ms	16.264 ms	🟢 1.6x faster

Windows

_{70b68ca083f6 · p50 latency, lower is faster. Full Raw Benchmarks}

Scenario	rmux	tmux (WSL)	zellij	vs tmux (WSL)
Create background session (default shell)	25.381 ms	86.820 ms	-	🟢 3.4x faster
List sessions	22.345 ms	92.790 ms	128.624 ms	🟢 4.2x faster
Split pane left/right	32.816 ms	89.223 ms	95.949 ms	🟢 2.7x faster
Split pane top/bottom	32.704 ms	90.214 ms	95.584 ms	🟢 2.8x faster
Send keys to pane/window	20.829 ms	87.857 ms	83.411 ms	🟢 4.2x faster
Capture visible output	21.019 ms	86.296 ms	82.153 ms	🟢 4.1x faster
Capture full scrollback	21.259 ms	89.079 ms	-	🟢 4.2x faster
List tabs/windows	19.537 ms	85.845 ms	78.786 ms	🟢 4.4x faster
Grow pane by 1 cell	21.172 ms	85.812 ms	-	🟢 4.1x faster
Terminate session	20.941 ms	88.374 ms	92.412 ms	🟢 4.2x faster

macOS

_{70b68ca083f6 · p50 latency, lower is faster. Full Raw Benchmarks}

_{macOS note: measured on a MacBook Pro. Most operations finish around 5 ms on this machine; read those rows as "same speed".}

Scenario	rmux	tmux	zellij	vs tmux
Create background session (default shell)	4.890 ms	9.526 ms	-	🟢 1.9x faster
List sessions	4.630 ms	4.564 ms	19.727 ms	≈ same speed
Split pane left/right	4.645 ms	4.580 ms	19.657 ms	≈ same speed
Split pane top/bottom	4.642 ms	4.576 ms	19.846 ms	≈ same speed
Send keys to pane/window	4.616 ms	4.536 ms	19.867 ms	≈ same speed
Capture visible output	4.617 ms	4.545 ms	19.497 ms	≈ same speed
Capture full scrollback	4.666 ms	4.568 ms	-	≈ same speed
List tabs/windows	11.410 ms	11.635 ms	19.845 ms	≈ same speed
Grow pane by 1 cell	4.685 ms	4.582 ms	-	≈ same speed
Terminate session	4.725 ms	4.623 ms	19.720 ms	≈ same speed

Reproducing

Run the scripts locally on each OS, gather the JSON files, and generate the Markdown:

scripts/bench/run-unix.sh --out target/benchmarks/linux.json
scripts/bench/run-windows.ps1 -Out target/benchmarks/windows.json
scripts/bench/run-unix.sh --out target/benchmarks/macos.json
python3 scripts/bench/render.py target/benchmarks/*.json --output docs/benchmarks.md

Each OS measures itself locally, so no number here includes SSH or remote-shell latency.

Note: Scripts run safely in isolated sessions and only kill their own test processes.

Methodology

The render.py script updates this Markdown file and linked CSVs based on the JSON results.

Each JSON artifact records the commit, platform, tool versions and every p50/p95 sample behind a cell.

-: Operation is unavailable or not comparable. Missing adapters and failed samples are omitted, never estimated.

Zellij and GNU screen: results only show exact equivalents or close approximations.

Tools reached through a compatibility layer are labelled as such, and the JSON records each measured tool's environment. On Windows the ratio compares RMUX against tmux under WSL, never against a native Windows multiplexer.

≈ same speed spans 0.80x to 1.25x, the spread repeated samples actually show on these hosts; a narrower claim would publish host noise as a result.

Generated at 2026-07-16T22:27:58Z from commit 70b68ca083f6.