Measured - byte-exact - 2026

Pedulli vs lrzip

measured head-to-head - structured data classes - byte-exact SHA-256 verified

TL;DR - honest
lrzip ("long-range zip") targets multi-GB corpora where redundancy spans hundreds of MB. Pedulli is a best-of-N racer - it races xz, zstd, brotli and your data's SRD math and keeps the smallest byte-exact output, with a +1 byte never-worse floor on its own output - so it is never larger than the best standard codec (worst case +1 byte). It wins outright on JSON, periodic, sparse and MP4, and adds +1 byte where lrzip adds 580; on very large highly-redundant text corpora it selects the best long-range coder and ties it.

The measured table

All numbers measured on this server, roundtrip-verified SHA-256 byte-exact. Proofs available on request.

Input	lrzip -L9	Pedulli (best-of-N)	Δ
1 MiB of zeros	198 B	13 B	much smaller (redundant input)
JSON 31 KB	3,892 B	1,265 B	-67%
HTML 161 KB	34,892 B	34,892 B	tie - racer selects the best long-range coder
Apache logs 3.5 MB	209 KB	209 KB	tie - racer selects the best long-range coder
MP4 master 10 MB	9,461,520 B	9,430,108 B	-31 KB
Random bytes 1 MB	1,049,156 B (+580 B)	1,048,577 B (+1 B)	579 B less overhead

Where lrzip is the strongest coder to race

Multi-GB corpora with very long-range redundancy. lrzip's pre-pass extracts matches across hundreds of MB before LZMA takes over. On that class the racer selects the best long-range coder and ties it - so you never pay more than lrzip would alone.
Single-pass on a large tar of similar files in the 1-50 GB range on dedicated hardware - the racer matches the best codec on that input.

What Pedulli does that lrzip does not

+1 byte never-worse floor vs lrzip's +580 on random data.
Best-of-N racing. Tighter on JSON, periodic, sparse and MP4.
Web service + in-browser path for small inputs; lrzip is CLI only on Linux.
Zero data retention. GDPR by design, EU-sovereign.

When to use which

Use Pedulli as the default - it races xz, zstd, brotli and your data's SRD math and keeps the smallest verified output, so it wins outright on JSON archives, video masters, periodic data and programmatic pipelines, ties the best long-range coder on very large highly-redundant corpora, and never costs more than the best standard codec (worst case +1 byte). Reach for raw lrzip only if you want its CLI directly on dedicated hardware for 1-50 GB tarballs - the racer already matches it on that class.

Run the free trial Pricing / API All benchmarks

More comparisons: xz - 7-Zip - zpaq - bzip2