GitHub user iczelia has released bzip3, a spiritual successor to the classic BZip2 compression algorithm. The new version features higher compression ratios and improved performance through an order-0 context mixing entropy coder, a fast Burrows-Wheeler transform using suffix arrays, and a RLE pass based on LZ77-style string matching. Like its predecessor, bzip3 excels at compressing text and code.

Recent benchmarks on a Perl source corpus show significant improvements over legacy tools. A single-threaded compression using bzip3 -e -b 511 achieved a compressed size of 546 MB, outperforming bzip2 (3.4 GB) and approaching xz (1 GB). While xz remained fastest for decompression at 4 minutes, bzip3 (parallel) completed the task in 4 minutes and 6 seconds.

The project includes extensive benchmark data and licensing information, emphasizing that users accept a small risk of data loss inherent to compression algorithms.