bzip3 wants to be the successor BZip2 never quite got

# bzip3 wants to be the successor BZip2 never quite got

*Event date: 2026-09-07*

A modest-looking GitHub project is trying to modernize one of the oldest names in file compression. bzip3, maintained by iczelia, describes itself as “a better and stronger spiritual successor to BZip2,” and says it offers higher compression ratios and better performance than its predecessor.

The project’s pitch is straightforward: keep the things that made BZip2 attractive for text and source code, then tighten up the implementation. According to the repository, bzip3 uses an order-0 context-mixing entropy coder, a fast Burrows-Wheeler transform implementation based on suffix arrays, and a run-length plus Lempel Ziv and prediction pass built on LZ77-style matching and PPM-style context modeling. That is a lot of compression jargon, but the headline claim is easy to understand: smaller files, faster decompression and better results on text-heavy archives.

The repository backs the claim with benchmark examples. It says the author downloaded every version of Perl 5, packed the results into one archive and compared compressors against that dataset. In those tests, the project presents bzip3 as a serious contender rather than a toy. The readme also points readers to other benchmark work comparing bzip3 against Turbo-Range-Coder, BSC and similar tools.

Bzip3 is not trying to hide from the practical side of compression either. The project notes that performance depends heavily on the compiler, and says x64 Linux builds with clang 13 can reach compression speeds as high as 17 MiB/s and decompression speeds of 23 MiB/s per thread. It adds that Windows and 32-bit builds can be slower, which is a reminder that low-level utility software still lives or dies on the details of the toolchain and target platform.

Like the older format it seeks to replace, the project says it is especially good at text and code. It also ships with installation guidance, including Homebrew instructions for macOS users, which suggests the developer wants the tool to be usable outside a benchmark chart. The code is licensed under LGPLv3 only, and the repository explicitly says it is not dual-licensed.

There is, however, a conspicuous warning in the project itself. The readme tells users not to compress anything unless they are prepared to accept the possibility, however small, that data may not be recoverable. That caveat is standard for serious compression software, but it is still a reminder that the promise of a stronger successor comes with a real-world burden: trust, testing and correct implementation matter as much as speed.

The news value here is not that bzip3 has displaced a mainstream standard; it has not. The interest is that the project is keeping an old-school compression idea alive in an era when cloud storage, ZIP files and proprietary platforms dominate most users’ attention. Projects like this matter because they remind the software world that infrastructure utilities can still improve in small but meaningful ways, especially for people who care about text archives, code bundles and long-term storage.

Bzip3 may stay niche. But if its claims hold up in broader use, it could become one of those quiet tools that people notice only after they need it, which is often how the best systems software earns its place.