Benchmarks
gopdfrab, veraPDF and PDFBox Preflight validating the same 773-file PDF/A-1b corpus on one machine.
| Metric | gopdfrab | PDFBox Preflight | veraPDF |
|---|---|---|---|
| Batch throughput files / sec · higher is better | 2,939 | 155 | 130 |
| Cold single-file latency ms · lower is better | 9.5 | 295 | 991 |
| Peak memory MB · lower is better | 80 | 945 | 728 |
| Deployment footprint MB · lower is better | 4.3 | 11.7 | 15.3 |
Batch throughput
files / sec · higher is betterWhole 773-file corpus in one process. gopdfrab finishes in 0.27 s.
Cold single-file latency
ms · lower is betterOne process per file, median corpus file, including process/VM startup.
Peak memory
MB · lower is betterMax RSS over the batch run.
Deployment footprint
MB · lower is betterThe CLI includes the fonts, colour profiles and renderer needed for PDF/A conversion. Both Java-based tools also require a JRE (+40 to 200 MB).
Methodology
One run of the benchmark harness in the gopdfrab repo (commit 1834292, 2026-08-04) over the combined Isartor + veraPDF PDF/A-1b corpora (773 files), on one Intel Core Ultra 7 255H running linux/amd64, against veraPDF 1.30.2 and PDFBox Preflight 3.0.7. Cold runs one process per file and includes VM startup, batch runs one process for the whole corpus and isolates the engine. Startup alone is 7.9 ms for gopdfrab against 102 ms for Preflight and 648 ms for veraPDF. Absolute numbers depend on hardware and other factors. Batch throughput measures the engine. Cold latency includes process startup, which is what a CI step or serverless call pays.
| Measured | 2026-08-04 |
|---|---|
| Hardware | Intel Core Ultra 7 255H, linux/amd64 |
| gopdfrab | commit 1834292, go1.26.5 |
| veraPDF | 1.30.2 |
| PDFBox Preflight | 3.0.7 |
| Corpus | 773 files (Isartor + veraPDF PDF/A-1b) |
Scope
This measures cost per validation, not coverage. veraPDF also covers PDF/A-3, PDF/A-4 and PDF/UA. Within PDF/A-1b and PDF/A-2b, gopdfrab passes 569/569 files of veraPDF's corpus and 204/204 of Isartor. The harness is in the gopdfrab repository.