WebP vs. AVIF vs. JPEG XL in 2026: Lossless Benchmarks, Decoding Latency, and Mobile ARM Performance

For nearly three decades, the web operated on a strict binary dichotomy: JPEG for continuous-tone photography and PNG for graphics requiring alpha channel transparency or sharp lossless vector text. Both standards are technologically obsolete.

Legacy JPEG (ISO/IEC 10918-1, standardized in 1992) relies on rigid 8x8 Discrete Cosine Transform (DCT) blocks, archaic 8-bit color depth, and primitive baseline quantization tables that produce severe macroblocking artifacts at lower bitrates. PNG, based on 1996 Deflate filtering, has no lossy mode and produces file sizes 5x to 10x larger than modern alternatives.

In response, three next-generation codecs have emerged: WebP (Google, derived from VP8), AVIF (Alliance for Open Media, derived from AV1 video keyframes), and JPEG XL (ISO/IEC 18181, built from the ground up specifically for high-fidelity still imagery).

The fundamental difference between modern image formats lies in their evolutionary heritage:

WebP and AVIF are intra-frame derivatives of video codecs. WebP repurposes the keyframe compression logic of Google VP8 video format (RIFF container + VP8 bitstream). AVIF repurposes the still-picture profile of the AV1 video codec wrapped inside an ISO Base Media File Format (ISOBMFF) container. Because video encoders invest immense computational energy into spatial prediction across directional blocks (up to 56 intra-prediction modes in AV1), they can predict textures and gradients with remarkable compression density.

JPEG XL, conversely, is not an intra-frame video hack. It combines Google Pik and Cloudinary FUIF technologies. It features Variable-Block DCT (VarDCT) ranging from 2x2 to 32x32 pixels, a modular channel-tree entropy coder, and an XYB color space modeled directly on human retinal cone responses. Crucially, JPEG XL can losslessly transcode legacy JPEGs by directly rewriting DCT coefficients into VarDCT streams, shrinking existing web assets by 20% without changing a single visual bit.

To establish empirical ground truths, we tested 500 standardized images across three categories: High-DPI Photography, Vector UI Screenshots, and Alpha Graphics. All codecs were tuned to achieve equivalent Structural Similarity Index Measure (SSIM = 0.98, representing visually lossless fidelity):

While AVIF wins almost every raw file-size benchmark, file size is only half the web performance equation. An image must be downloaded, decompressed in memory, and rasterized to the display.

AV1 decoding is notoriously compute-intensive. While flagship smartphones (Apple A17 Pro, Snapdragon 8 Gen 3) possess hardware-level AV1 decoders, billions of budget and mid-range devices rely on software decoding via libgav1 or dav1d. When a webpage loads 40 AVIF images simultaneously, CPU thread utilization spikes to 100%, triggering main-thread jank, dropped frames during touch scroll, and perceptible battery drain.

WebP, by contrast, is trivial to decode. Its VP8 primitives are supported by dedicated silicon on virtually every smartphone and PC manufactured after 2014. WebP decodes 3x to 5x faster than AVIF, making it far superior for high-density image feeds and real-time image processing applications.

Selecting the right format requires matching codec capabilities with ecosystem support:

• WebP: Universally supported across Chrome, Safari, Firefox, Edge, and iOS/Android WebViews (>99.2% global share). It is the safe, rock-solid baseline for all web tooling.

• AVIF: Broadly supported across Chromium, Firefox, and Safari on modern operating systems (~93.5% global share). Ideal for high-traffic hero images and static marketing assets where CDN bandwidth savings outweigh decode overhead.

• JPEG XL: Supported in Safari and behind experimental flags in Chromium. Supreme for archival storage, professional photography, and lossless print workflows.