Measured, and said
A number without its method is a claim, so the method stands beside each. Where vobuda loses to a terminal written for speed, the cause is written here rather than hidden.
What was measured, and on what
One machine, one run of the program's own measuring script, which launches it on a settings folder of its own and asks it through the command line.
- Version
- 4.2.2, the release bundle
- Date
- 15 October 2026
- Machine
- arm64, Apple M4, 34 GB, macOS 27.0.1
- Method
- The program launched on a settings folder of its own, and everything read back through the command line
Where it loses, and why
Vobuda is a Tauri program: a Rust core and a WebKit window drawing xterm.js. That costs memory that a GPU terminal written in Zig or Rust does not pay, because a web view is a browser. It is also what buys the interface in every language it ships, the pages, the editor and the panels. A terminal written only for speed will use less memory than this, and we do not claim otherwise.
We have not run the others on the same day on the same machine, so this page names no figure for them. Their own sites, and reviewers who measured them, are the authority for what they use.
Memory
Resident size of the program's own process plus the WebKit processes that appeared with it, as Activity Monitor counts them, three seconds after the first state was written.
| What | MB |
|---|---|
| An empty window, one shell | 323 |
| After 200 000 lines of output | 496 |
Speed
Timed from outside, through the command line, on a release build.
| What | Value | How it was timed |
|---|---|---|
| Start | 0.51 s | From the launch to the first state written |
seq 1 200000 through a run block | 2.2 MB/s (0.58 s for 1.3 MB) | The whole road: pty, parser, screen, from the shell's start to the exit line the block's parser read. Not the renderer alone, which is what a benchmark of a terminal usually times |
A command's round trip, vobuda run true | 458 ms | Median of five |
| Output gathered before sending | 0 ms | Read off the block |
Input latency is not measured here. It needs an instrument that watches the screen, and none of ours does. The round trip above is the nearest honest number.
What the program does
The classical list a programmer checks in the first hour.
| What | Vobuda |
|---|---|
| Search over the whole scrollback | Yes. Cmd+F searches the block's buffer, up to the scrollback setting (10000 lines by default) |
| Ligatures | Yes, through xterm's ligatures addon, off by default (term.ligatures; the shipped font, Menlo, has none) |
| Images in the terminal | Sixel and iTerm2 inline images through the image addon. The Kitty graphics protocol is not supported: the addon does not speak it |
| Prompt marks (OSC 133) | Yes, for zsh, bash and fish |
| Notifications (OSC 9, 99, 777) | Yes |
| GPU rendering | Yes, xterm's WebGL renderer, with a DOM fallback for glyphs the font lacks |
What the numbers are not
One machine, one run, on the build named above. Not a benchmark suite, and not a comparison run on the same day against the others. The figures are redone for each release, and the date and version at the top say which release these are.
Memory and speed are the parts of a program that shows itself to anyone. What it does to your machine outside its window is written on its own page.