Ghostty vs Kitty (2026): Speed, Scripting, and the Cost of Switching

Most Ghostty vs Kitty comparisons stop at the benchmark table, and the table is real: in our May 2026 run on a MacBook Pro M3 Max, Ghostty opened a cold window in 68 ms and Kitty took 145 ms. You will stop noticing that inside a week. What you notice later, sometimes a year later, is how much it costs to leave. Pick Ghostty if your terminal is a window you put a shell and tmux into. Pick Kitty if you script the terminal itself or need BSD. If you honestly can't tell which of those describes you, start on Ghostty. It is the cheaper one to walk away from, and the rest of this page explains why that matters more than 77 milliseconds.

Ghostty vs Kitty benchmarks

These numbers come from our Ghostty vs iTerm2 vs Kitty benchmark: fresh installs, default settings, macOS 15.6, three runs averaged.

Measure	Ghostty	Kitty
Cold startup	68 ms	145 ms
Idle RAM	24 MB	42 MB
Keyboard latency	1.2 ms	1.8 ms
Scrollback render	144 fps	144 fps
10K line reflow	12 ms	18 ms
grep through 1M log lines	1.4 s	1.3 s
Platforms	macOS, Linux	macOS, Linux, BSD
Renderer	Metal on macOS, OpenGL on Linux	OpenGL
Scripting	None, by design	Kittens (Python), remote-control socket
License	MIT	GPL 3

Ghostty takes four of the six speed rows, ties one, and loses the grep by a tenth of a second. The latency gap (0.6 ms) sits under what most people can perceive. Kitty also runs one backend process for all its windows, so after the first launch new windows open almost instantly, which means the startup difference mostly shows up once a day, right after you reboot. If you run eight windows, the memory gap is about 144 MB. That's the whole performance story.

What Kitty can do that Ghostty won't

Kitty is a programmable terminal. Kittens are small Python programs that ship inside it: kitten ssh, kitten diff, kitten icat, kitten themes, kitten hints. Turn on allow_remote_control and any shell script can open tabs, type into panes, or repaint the window from outside:

# kitty.conf
allow_remote_control yes
listen_on unix:/tmp/mykitty

# amber background while you are on prod
kitty @ --to unix:/tmp/mykitty set-colors background=#3a2a12

Add named layouts (tall, fat, grid, stack) that you cycle from the keyboard and session files that boot a whole workspace with kitty --session, and you have a terminal that people build tooling on. Ghostty has no equivalent for any of it. Mitchell Hashimoto's design bet is that tmux already handles sessions and your shell already handles scripting, so the terminal should be a very fast surface and stop there. Nobody should expect that bet to change in a point release.

Where Ghostty wins

On a Mac, Ghostty is a SwiftUI app drawing through Metal, with real macOS tabs and System Settings integration. Kitty on macOS goes through OpenGL, which Apple has deprecated; it works fine and still reads a little like a Linux program that moved in. Ghostty's config is one plain key = value file at ~/.config/ghostty/config, and for most developers the whole thing is under fifteen lines. The MIT license matters at exactly one kind of company, the kind where GPL 3 triggers a legal review, and there it matters a lot.

The switching-cost test

None of the top-ranking Ghostty vs Kitty comparisons we read cover this, and it is the part we'd weigh most. Config migration between these two is lopsided. Every line of a Ghostty config has a Kitty counterpart. The reverse is false.

In your config	Moving Ghostty to Kitty	Moving Kitty to Ghostty
Fonts, colors, padding, cursor	Line for line	Line for line
Keybindings	keybind becomes map	map becomes keybind
enabled_layouts	n/a	No equivalent, hand-placed splits
startup_session	n/a	No equivalent, no session files
allow_remote_control and kitty @ scripts	n/a	No equivalent, no socket
kitten ssh	n/a	Manual terminfo install per host

So the order you try them in changes the price. Ghostty first, then Kitty if you outgrow it, costs an afternoon of renaming keys. Kitty first, then Ghostty a year later, can cost every script in your dotfiles that starts with kitty @, and lock-in like that accumulates without anyone deciding on it, one small script at a time. Before you decide, run this:

grep -rnE "kitty @|kitten |enabled_layouts|startup_session" ~/.config ~/dotfiles 2>/dev/null

Zero hits means you can switch to Ghostty for free and get back about 77 ms of startup and 18 MB per window. A page of hits means Kitty has become part of your toolchain, and you should probably stay put.

SSH is where Ghostty bites

Ghostty sets TERM=xterm-ghostty. On a server that has never seen that terminfo entry you get tput: unknown terminal errors and arrow keys that print escape codes in less. Full-screen TUIs draw in the wrong place too. Kitty sidesteps this because kitten ssh copies its terminfo over before your shell starts. With Ghostty you do it yourself, once per host:

infocmp -x xterm-ghostty | ssh user@server -- tic -x -

Or you fall back to TERM=xterm-256color and lose extended key reporting and styled underlines. If your week involves SSHing into dozens of brand-new machines, this one chore is a fair reason to stay on Kitty regardless of anything in the benchmark table.

Inside Neovim

You won't be able to tell them apart in Neovim, because Ghostty implements both of the protocols Kitty invented, the keyboard protocol and the graphics protocol, so undercurl diagnostics and image.nvim work the same in each.

Kitty vs Ghostty for current Kitty users

Run the grep above. If your kitty.conf is fonts and colors plus a few map lines, the move is mostly renaming keys. If you're on BSD, stop here, Ghostty doesn't run there. Kitty has also been packaged in mainstream Linux distributions since 2017, while Ghostty's Linux build is a newer GTK4 app whose packaging depends more on your distro. On Linux, Ghostty also loses its biggest Mac advantage, since you're comparing two toolkit apps instead of a native app against a port.

Ghostty vs iTerm2

The same leave-cost lens works here, and it points the other way. Against iTerm2, Ghostty is far faster: 68 ms against 312 ms to open, 24 MB against 78 MB idle, 1.2 ms of keyboard latency against 4.1 ms, and 144 fps scrollback against iTerm2's 60 fps cap. But iTerm2 users tend to depend on things Ghostty does not attempt, like tmux -CC control mode that turns tmux windows into native tabs, triggers that color errors as a log scrolls past, automatic profile switching when you SSH into prod, and Cmd+F regex search over scrollback. Open iTerm2's settings and count how many of those you touched this month. If the answer is none, switch. Our Ghostty vs iTerm2 Mac comparison walks the iTerm2 vs Ghostty decision setting by setting.

Ghostty vs Alacritty

Alacritty is the purist's pick. No tabs, no splits, no inline images, no ligatures (the maintainers have declined to add them). It expects tmux or your window manager to do the multiplexing, it is configured in TOML, and unlike Ghostty it runs on Windows. In our five-terminal roundup, a separate test from the table above, Alacritty idled at 22 MB to Ghostty's 28 MB. If you need one config across macOS, Linux and Windows, or you already live in tmux and want the terminal to do nothing else, Alacritty is still a good answer. If you use Fira Code or JetBrains Mono ligatures, it isn't. The four-way Mac showdown and the best terminal emulators of 2026 roundup cover Alacritty and Warp in more depth.

Verdict

Ghostty, for most developers on a Mac who run tmux or Zellij and want the terminal to disappear. Kitty, the moment you catch yourself writing a script that talks to your terminal, or if BSD or constant fresh-host SSH is part of your job. Both are free. Whichever you choose, keep sessions in a multiplexer (our tmux vs Zellij comparison covers that layer) and your terminal stays a replaceable part, which is the whole point of the grep test.