Jake Archibald has done what the format wars needed: a meticulous, multi-encoder, multi-metric benchmark comparing AVIF and JPEG XL across every web-relevant scenario. The verdict is clear-eyed and unsentimental. For lossy web images — the vast majority of what any site serves — AVIF produces smaller files, encodes faster, decodes faster, and uses less memory. The gap ranges from modest (6–15% at max effort for photos) to enormous (30–90% for screenshots and illustrations at default effort). JPEG XL's best moments are outliers, not the norm. The piece is careful about encoders in a way most comparisons aren't. It tests libjxl (the reference JPEG XL encoder), jixel (a newer alternative), and avifenc, acknowledging that encoder quality matters as much as format spec. jixel closes the gap with AVIF on some photo benchmarks, but only at punishing encode times and memory costs. At equivalent compute budgets, AVIF still wins. libjxl's higher effort settings are particularly damning — slower, hungrier, and sometimes producing larger files than its own defaults. Transparency handling is a blowout. AVIF encodes alpha channels efficiently in lossy mode; libjxl's files are 31–70% larger depending on effort, and jixel encodes transparency losslessly by default, ballooning files to 2–3x AVIF's size. For screenshots and illustrations — flat color, sharp edges, text — AVIF's advantage grows to the point where JPEG XL is barely competitive. JPEG XL has two genuine wins. Lossless compression is clearly superior: lossless WebP is 9–17% larger for photos, lossless AVIF 30–120% larger. And JPEG XL's lossless JPEG transcoding — shrinking existing JPEGs by 9–20% without any quality loss — is a unique capability no other format offers. But Archibald is honest that lossless images are almost never what you want on the web, making this a niche advantage. The progressive rendering comparison is surprisingly nuanced. JPEG XL's progressive mode via jixel is nearly free at high effort, but AVIF has its own progressive rendering in Chrome (with Firefox support coming), offering more control over preview quality and faster decode. The HDR story is muddier — JPEG XL lacks gain map support in browsers despite the format supporting it, while the two quality metrics used disagree on which format handles non-gain-mapped HDR better. Archibald's practical recommendation lands cleanly: default to AVIF 4:4:4 at the highest effort you can tolerate, and try JPEG XL via jixel only when AVIF produces an unusually large file. He explicitly warns against trusting AI-generated summaries of this topic, noting widespread misinformation. The methodology is transparent — specific encoder versions, a 2024 MacBook Pro, SSIMULACRA2 as the primary metric, full data tables published — and the writing trusts readers to examine the evidence themselves. This is the kind of technical comparison that actually moves a field forward: no format tribalism, no spec-sheet theater, just measured results with named tools and reproducible conditions. Both formats have improved rapidly, which means the recommendation could shift. But today, the data is the data.
