How we measured this
On August 15, 2026, we ran the same 22 real .pub files (17 Apache POI test files, 5 real-world Bugzilla attachments, spanning Publisher 98 through 2010) through stillPUB's converter, LibreOffice Draw 26.2.5.2, and Scribus 1.6.6. Structural recovery was counted, not eyeballed: pages, text frames, shapes, tables and paragraphs, read out of stillPUB's JSON and LibreOffice's flat ODF export with an XML parser. No pixel diffing was done. The full method, raw counts and per-file appendix are in the benchmark writeup; this page summarizes it and adds nothing it did not measure.
Structural fidelity: an exact tie with LibreOffice
On this corpus, stillPUB does not out-parse LibreOffice Draw, and we are not going to pretend it does. Page counts, frame and shape counts, table counts, and paragraph counts, including the empty ones, are identical between stillPUB and LibreOffice on all 22 files, once stillPUB's metadata-parse patch is applied (before the patch, stillPUB failed on 2 of the 22 files; the patch brings it up to LibreOffice's 22 of 22, not past it). Both tools sit on the same underlying parser, libmspub, so whatever it recovers, both get, and whatever it drops, such as WordArt reduced to bare shapes or hyperlink targets replaced with a literal "#", both drop the same way. If a comparison page tells you one tool "beats" the other on raw import fidelity here, it is not reading its own numbers.
Comparison table
| Measure | stillPUB | LibreOffice Draw | Scribus |
|---|---|---|---|
| Structural recovery, 22-file corpus | 22 of 22 files, tied with LibreOffice | 22 of 22 files, tied with stillPUB | Partial: 2 of 6 attempted files opened before the run was time-boxed |
| Runs where | In the browser, no install | Desktop application, install required | Desktop application, install required |
| File upload required | No; the file is read on your own computer | No; it is a local desktop app | No; it is a local desktop app |
| Per-file degradation reporting | Explicit warnings list naming what did not survive | None; conversion is silent | None; a hang gives no error at all |
| Document model after import | Editable, purpose-built, fonts labeled inline per span | Editable in Draw; correct ODF, but styles are indirected through named style references | Editable Scribus canvas, on the files that open |
| Print-ready PDF export | Yes | Yes, a mature pipeline, 22 of 22 files in the benchmark | Yes, on the files that opened |
| Page-image export (JPG/PNG) | Yes | Not part of its Publisher workflow | Not part of its Publisher workflow |
| Per-span language tags | Dropped (ICU stub); noted by its own warnings | Kept | Not verified |
| Embedded WMF images | At the August benchmark, kept as raw bytes with a warning. Since v0.3.3 the browser editor decodes WMF to PNG on import; the headless command-line export still skips WMF and says so | Converted to PNG, silently | Not verified (the one corpus file with heavy WMF content hangs its importer) |
| Reliability on real-world files (not developer samples) | Not separately isolated in the benchmark; developer and real-world files were both 22 of 22 | Not separately isolated in the benchmark; developer and real-world files were both 22 of 22 | 4 of 5 real-world Bugzilla files hung the importer for a full 180 seconds each, with no error |
Where stillPUB is genuinely ahead
- An honest warnings channel. Every stillPUB conversion carries a list of what did not survive, by name. LibreOffice converts the same file with no message at all; you find out what broke by comparing pages yourself.
- A document you can keep editing, with fonts labeled. stillPUB's document model resolves each span's font and style inline instead of pointing at a named style you would have to look up, and it is built to be edited in a Publisher-shaped interface rather than a general drawing canvas.
- No install, no upload. stillPUB runs in a desktop browser tab. There is nothing to download and nothing sent to a server; both LibreOffice and Scribus are installs.
- Print-ready PDF and page-image export. stillPUB exports a print-ready PDF and JPG or PNG page images from the same file, without leaving the browser.
Where LibreOffice is genuinely ahead
These are the benchmark's own findings, and they stand as written:
- Per-span language tags. LibreOffice keeps them; stillPUB's ICU stub drops them, affecting more than a thousand text spans across 21 of the 22 files, by stillPUB's own reporting.
- Renderable images from legacy metafiles. At the time of the benchmark LibreOffice converted embedded WMF images to PNG on the way in and stillPUB kept the raw bytes and warned. stillPUB's browser editor has decoded WMF on import since v0.3.3, so this gap now applies only to its headless command-line export, which still skips WMF images and reports each one.
- A mature, shipping print pipeline. LibreOffice's command-line PDF export produced a non-empty PDF for all 22 files at roughly a second each once warmed up. This is years of polish stillPUB's PDF export does not yet carry.
Scribus: partial results, time-boxed
Scribus 1.6.6 installs cleanly and is scriptable, but the benchmark run on it was stopped after 6 of the 22 files. Four of the five real-world Bugzilla documents hung its importer for the full 3-minute watchdog each, with no error message, no partial document and no way to tell from the outside what had gone wrong. The remaining 16 developer-sample files were not attempted, because burning roughly 12 minutes on hangs before reaching them was not a reasonable use of a benchmark session. The 2 files that did open showed the same libmspub-level degradation as stillPUB and LibreOffice, with a small amount of extra object loss on one file. Where Scribus works, it is a real full desktop publishing canvas with fast PDF export. On measured behavior, it is not a dependable way to open a folder of real Publisher files.
Which should you use
If you already have LibreOffice installed and do not need a Publisher-style editing interface, LibreOffice Draw is a fine free choice. It ties stillPUB on structural recovery, keeps per-span language tags, and has a mature PDF export pipeline. You trade that for a desktop install, a generic drawing canvas instead of a page-layout tool, and no warning if something silently failed to convert.
If you are a designer who wants a full desktop publishing tool and are prepared to work around occasional import hangs, Scribus is worth having installed, particularly for new work rather than as your primary way of opening old .pub files.
If you want to open a .pub file right now, in the browser, with no install, no upload, and a plain list of anything that did not survive, that is what stillPUB is built for.
Caveats: what this benchmark did not measure
- Visual fidelity. No pixel-diffing was done against reference renders or real Publisher output. Identical object counts do not prove identical layout, line breaking or font substitution.
- Editing UX. Nobody edited anything in the benchmark run. "Editable" means the structure survives import, not that the editing experience was compared.
- stillPUB's own time-to-print was not separately measured on the CLI at the time of the benchmark; its converter timing is a stated proxy, not a print-pipeline measurement.
- Corpus skew. 17 of the 22 files are developer test samples rather than real-world documents, and all files are Publisher 98 through 2010; nothing here says how any of these tools behaves on Publisher 2013 through 2016 output, or on CJK or right-to-left documents.
- Scribus coverage is partial. Only 6 of 22 files were attempted before the run was time-boxed; the 4 importer hangs were not root-caused.
- Univik and Megal were not benchmarked at all. Both appear on other Publisher-alternative lists, including our own broader comparison page, but neither was run against this corpus, so this page makes no fidelity claim about either one. Do not read their absence from the table above as a ranking.
The benchmark this page is built on lives in the project's own documentation and is reproducible from the method described above; nothing on this page goes beyond what it measured.