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: ```bash 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`.