Benchmarks are easy to game. Cherry-pick your test files, choose a weak baseline, ignore edge cases, and you can make anything look good. We wanted to do the opposite.

Methodology

Our benchmark suite uses:

Baseline: LZMA-9 (maximum compression level) — the strongest general-purpose compressor
Also tested: gzip-9, bz2-9, brotli-11, zstd-19
Files: 3,184 real-world files across 20 file types
Verification: Byte-exact round-trip on every file
Guarantee: Never-worse — PZIP output ≤ LZMA output always

Results by File Type

Our strongest results come from structured data:

CSV: Up to 68.8% smaller than LZMA — column types, dictionary encoding, and delta compression combine powerfully
JSONL: Up to 93.5% smaller — schema detection and key dictionary extraction
Log files: Up to 86.4% smaller — PZIP wins on structured/repetitive logs
XLSX: Up to 84.7% smaller — OOXML unpacking, shared string optimization, style factoring
WAV Audio: Up to 47% smaller — sample-level prediction and multi-codec gateway

Even already-compressed formats show gains:

PDF: Up to 49% smaller — stream recompression and metadata optimization
JPEG/PNG: Wins via header extraction and metadata separation — never worse

All Test Files Are Downloadable

Every file in our benchmark suite is available for download. You can verify our results yourself. No trust required — just math.

See the full results at pzip.net/benchmarks.