A founder's benchmark of 42 compression formats favors Zstd and LZMA2

A detailed benchmark of 42 formats reveals a critical flaw in trusting exit codes and provides a clear framework for choosing between Zstandard's speed and LZMA2's ratio for modern applications.

The Answer Up Front

For developers shipping software or data where decompression speed is critical, Zstandard is the correct default choice. For personal, long-term archiving where achieving the absolute best compression ratio is the goal and decompression is infrequent, 7-Zip using LZMA2 remains the top mainstream pick. The benchmark's most important contribution, however, is not a single winner but a methodological warning: verify archived content byte-for-byte, because a tool’s successful exit code is not a guarantee of data integrity.

Methodology

This v0 review analyzes a benchmark published by Andrew Dyster, founder of the browser-based archive tool ezyZip, on July 10, 2026. The source is a single blog post, and all performance numbers are the founder's published claims; independent verification is pending. Our analysis covers the benchmark's setup, its test corpus, its verification process, and its key findings as presented in the source article at https://dev.to/andrew_dyster_c88ccbb5180/i-benchmarked-42-compression-formats-spanning-four-decades-heres-what-to-actually-use-143o.

The test rig measured 42 distinct formats. Each was tested at its default and single "maximum compression" setting, deliberately avoiding complex tuning to reflect typical user behavior. The corpus was a 55 MB mix of text, office documents, images, and video, designed to mimic a realistic user folder. A critical and laudable feature of the methodology was its verification process. Instead of trusting program exit codes, every created archive was extracted and its contents were SHA-256 hashed against the original files. This review does not cover long-term performance, memory usage, or the impact of multi-threading, which the benchmark explicitly excluded.

What The Benchmark Shows

A realistic corpus matters

Most academic compression benchmarks use standardized corpora like Silesia or Canterbury, which are heavy on text and uncompressed data types. Dyster’s benchmark deliberately used a mixed-media corpus including JPEGs, H.264 video, and Microsoft Office documents. This is a more realistic test for general-purpose archiving. Office documents (.docx, .pptx) are themselves ZIP archives, so testing against them reveals how well a format handles already-compressed data. Similarly, including nearly incompressible media files establishes an honest baseline for performance, preventing skewed results from easily-squashed text files.

Byte-verification is non-negotiable

The most significant finding was a practical demonstration of silent data corruption. An older ARC implementation from 1985 produced an archive that its own extractor unpacked with a success exit code, despite printing a CRC warning and writing a file 55 bytes larger than the original. Had the benchmark relied only on exit codes, this critical failure would have been recorded as a success. The author's insistence on byte-for-byte content verification using SHA-256 hashes is the single most important lesson from this work. It's a stark reminder that a tool's self-report cannot be the final word on data integrity.

The Zstd vs. LZMA2 trade-off

The results table highlights the central conflict in modern compression: speed versus ratio. At maximum settings, 7-Zip (using LZMA2) achieved a 59.96% ratio, taking 4.43 seconds to compress and 0.43 seconds to decompress. Zstandard was close on ratio (61.33%) and slightly slower to compress (6.82s), but its decompression was nearly instant at 0.05 seconds. That is an 8.6x speedup in decompression for a negligible 1.37 percentage point sacrifice in compression ratio. This makes the choice clear: for any application where the user is waiting for decompression (like software installation or loading game assets), Zstandard is superior.

What's Interesting / What's Not

The interesting part of this benchmark is its pragmatic, opinionated methodology. It’s built to answer a practical question (“what should I zip my folder with?”) rather than an academic one. The focus on end-user tools, default settings, and especially the rigorous, content-aware verification process provides a template for how to conduct useful tool evaluations. The story of the silently corrupting ARC tool is a perfect, concrete example of why this matters.

What’s not interesting is the top and bottom of the performance table. It is no surprise that ancient formats like compress are poor, or that esoteric, context-mixing formats like zpaq achieve the best ratio (56.11%) at an impractical speed (144 seconds to compress). The actionable results are in the middle, where the trade-offs between mainstream tools like Zstandard and 7-Zip become clear. The benchmark correctly focuses the reader's attention there.

Pricing

The benchmark itself is a freely available article. The primary tools it identifies as top contenders (Zstandard, 7-Zip, xz) are open-source and available at no cost. Pricing snapshot taken July 10, 2026.

Verdict

This is an excellent example of a practical, reproducible benchmark that provides clear, actionable advice. Its primary value is its methodology. The insistence on a realistic corpus and byte-level verification over exit codes should be adopted as a standard for any storage or archival tool testing. Based on the author's reported data, the recommendation is straightforward. If you are distributing data and your users' time is a factor, use Zstandard; its phenomenal decompression speed is worth the minor cost in compression ratio. If you are creating a personal archive for long-term storage where size is the only metric that matters, 7-Zip (LZMA2) is the best-in-class choice.

What We'd Test Next

A v2 of this benchmark would benefit from measuring memory consumption for both compression and decompression, as this can be a limiting factor on constrained systems. We would also test performance on more specialized corpora, such as a large codebase or a folder of JSON logs, to see how the rankings shift. Finally, exploring the impact of multi-threading options would be valuable, as the original benchmark was deliberately single-threaded. This would better reflect performance on modern, multi-core server and desktop hardware.

The investor read

This benchmark underscores that foundational technologies like data compression remain a critical and evolving field. The clear performance win for Zstandard, a project originating from Facebook, demonstrates how hyperscaler investment in open-source tooling continues to shape infrastructure choices for the entire industry. For investors, the key signal is that raw performance metrics (like compression ratio) are often less important than user-experience-aligned metrics (like decompression speed). Startups building tools where data transfer or loading times are a factor must choose dependencies that optimize for the user's wait time. The author's company, ezyZip, also provides a model for bootstrapped growth: building authority and trust via rigorous, transparent technical content is a proven marketing strategy in the developer tool space.