H.264 vs H.265 vs AV1: Video Codec Comparison (2026)
What These Three Codecs Actually Are
H.264 (AVC), H.265 (HEVC), and AV1 are video compression standards — algorithms that decide how to shrink raw video data into a file you can actually store and stream. Raw 1080p60 video produces roughly 12 GB per minute. A codec's job is to throw away information your eyes won't notice while preserving what they will. H.264 was standardized in 2003 and became the backbone of everything from Blu-ray discs to YouTube uploads. It's backed by a patent pool managed by Via LA, which historically meant licensing fees for commercial encoders and decoders — though enforcement has been inconsistent. H.265 arrived in 2013, promising roughly 40–50% better compression than H.264 at equivalent visual quality. Its royalty situation became a mess: multiple competing patent pools (HEVC Advance, Via LA, Sisvel) created enough legal uncertainty that Apple, Google, and Mozilla initially refused to support it in browsers. AV1 is the royalty-free answer to that mess. Developed by the Alliance for Open Media — a coalition that includes Google, Apple, Netflix, Amazon, and Intel — AV1 was finalized in 2018 and targets 30% better efficiency than H.265. Because it's royalty-free, every major browser now supports it natively, and hardware decode support has become mainstream in devices released after 2022. Understanding this background matters because codec choice isn't purely technical. Licensing costs, hardware support, and encode speed all feed into the decision alongside pure compression efficiency.
Compression Efficiency: The Numbers That Actually Matter
Codec comparisons often get hand-wavy about quality differences, so here are concrete benchmarks from published research you can verify. Netflix's 2020 large-scale codec comparison tested all three at matched VMAF scores (a perceptual quality metric). At VMAF 93 for 1080p content, H.265 required roughly 45% less bitrate than H.264, and AV1 required about 35% less bitrate than H.265 — or roughly 60–65% less than H.264 overall. In practical terms: a 90-minute movie encoded at acceptable streaming quality (VMAF ~85) might run 4 GB in H.264, around 2.2 GB in H.265, and roughly 1.5 GB in AV1. For a streaming service delivering millions of views, that difference is enormous. For someone archiving home videos, it's less critical. Those gains aren't free. AV1 encodes at software speeds that are 10–20x slower than H.264 at comparable quality settings. Using the libaom-av1 encoder at its default 'good' speed preset (cpu-used=4) on a modern 8-core CPU, expect roughly 15–25 fps for 1080p content. H.264 with libx264 at medium preset encodes the same content at 150–300 fps on the same hardware. SVT-AV1, the encoder developed by Intel and Netflix, closes much of that gap. At preset 6, SVT-AV1 achieves 80–120 fps for 1080p while still beating H.265 on compression. It's the encoder CocoConvert uses for AV1 output, which is why our AV1 conversion jobs finish in a reasonable timeframe rather than hours. H.265 sits between the two in encode speed. libx265 at medium preset runs at roughly 30–60 fps for 1080p — slower than H.264 but far faster than libaom-av1.
Browser and Device Support in 2026
Support has shifted dramatically over the past three years, and the old rule of thumb — 'use H.264 for maximum compatibility' — now requires more nuance. H.264 remains universally supported. Every browser, every smart TV, every phone manufactured in the last decade plays it. If you're distributing video to an unknown audience with unknown devices, H.264 is still the safe default. H.265 has broad hardware support but inconsistent software support. iPhones since the iPhone 7 (2016) decode HEVC in hardware. Android support varies by chipset — Snapdragon 835 and later handle it fine, but budget MediaTek chips from 2019–2021 often don't. On desktop, Safari plays HEVC natively; Chrome and Firefox require hardware decode support and the right OS-level codec installed. On Windows, you may need to purchase the HEVC Video Extensions from the Microsoft Store ($0.99) for software decode fallback. This fragmentation makes H.265 awkward for web delivery. AV1 has become the practical sweet spot for web streaming. Chrome, Firefox, Edge, and Safari (since 16.4) all support it. Hardware decode is available on Apple Silicon Macs, Intel 11th-gen and later, AMD RDNA2 and later, Nvidia RTX 30-series and later, and virtually all Android devices with a 2022 or newer SoC. YouTube defaults to AV1 for most desktop streams. Netflix uses AV1 for mobile and select titles on desktop. For CocoConvert users: if you're converting video for a website you control and your analytics show a modern audience (Chrome/Safari on recent hardware), AV1 is a legitimate choice. If you're producing files for a corporate intranet with locked-down Windows 7 machines, H.264 is not optional. One honest caveat: CocoConvert's AV1 output targets web delivery. We don't currently support AV1 in MP4 containers with Dolby Vision HDR metadata, which some professional workflows require. For that use case, H.265 in an MKV or MP4 container remains the practical option.
Which Codec to Choose for Common Use Cases
Rather than a generic recommendation, here's how to think through specific scenarios. **Uploading to YouTube or social platforms:** Encode in H.264 at a high bitrate (10–20 Mbps for 1080p, 35–68 Mbps for 4K per YouTube's own guidelines) and let the platform re-encode. YouTube will produce AV1, VP9, and H.264 versions automatically. Sending AV1 to YouTube doesn't hurt, but it doesn't help either — they'll transcode it regardless. **Archiving raw footage:** H.265 is the practical choice. It's supported by every major NLE (Premiere Pro, DaVinci Resolve, Final Cut Pro), offers significant space savings over H.264, and doesn't have AV1's encode-time penalty. A 10-bit H.265 file in an MKV container at CRF 18 is a reasonable archival format for content you want to keep but not edit frequently. **Self-hosted web video:** AV1 at CRF 32–38 (SVT-AV1 scale) with H.264 as a fallback using the HTML5 `<video>` source element. This covers effectively 100% of browsers while giving modern users better quality at lower bandwidth. **Sharing with family or non-technical recipients:** H.264 in an MP4 container. No exceptions. It will play on their Samsung TV, their Windows laptop, their iPad, and their five-year-old Android phone without any codec installation prompts. **4K HDR content:** H.265 with HDR10 metadata is the most widely supported option. AV1 with HDR10 works in Chrome and Firefox but not universally in standalone media players. Dolby Vision requires H.265 or H.264 in specific container configurations — AV1 + Dolby Vision is a 2025-era format that hardware support hasn't caught up to yet. When using CocoConvert, you'll find these presets mapped to output profiles: 'Web Compatible' maps to H.264 Baseline/Main profile, 'Modern Web' uses AV1 via SVT-AV1, and 'Archive Quality' defaults to H.265 Main10 for HDR-capable sources.
Quality Settings That Actually Make a Difference
Codec choice matters less than quality settings in many real-world scenarios. A poorly configured H.265 encode will look worse than a well-configured H.264 encode at the same file size. All three codecs use a rate control parameter called CRF (Constant Rate Factor) for quality-based encoding. Lower values mean higher quality and larger files. The scales differ between codecs: - H.264 (libx264): CRF 18 is visually lossless, CRF 23 is the default (good quality), CRF 28 is acceptable for low-bandwidth delivery - H.265 (libx265): CRF 24 is roughly equivalent to H.264 CRF 23 in perceived quality, CRF 28 is a common streaming target - AV1 (SVT-AV1): CRF 35 is roughly equivalent to H.264 CRF 23; the scale runs 0–63 Beyond CRF, the encoding preset matters enormously for efficiency. libx264 at 'veryslow' preset produces files 10–15% smaller than 'medium' at the same CRF — useful for archival encodes where you run it once. For batch conversions where time matters, 'fast' or 'faster' is a reasonable tradeoff. For H.265, always specify `-tag:v hvc1` when outputting to MP4 containers if your target audience includes Apple devices. Without this tag, QuickTime and iOS will refuse to play the file even though the codec is supported. This is a common gotcha that trips up automated converters — CocoConvert applies this tag automatically for H.265 MP4 output. For AV1, the `--film-grain` parameter in SVT-AV1 (values 0–50) synthesizes grain at decode time rather than encoding it, which saves significant bitrate for noisy or grainy source material. Film content benefits from values around 8–12; clean digital sources should leave it at 0. One setting that's frequently overlooked: two-pass encoding. For fixed-bitrate delivery (streaming at a specific bitrate cap), two-pass encoding distributes bits more intelligently than single-pass VBR, typically improving quality by a visible margin at the same average bitrate. CocoConvert's 'Streaming Optimized' output profiles use two-pass for this reason.
Hardware Encoding: Speed vs. Quality Tradeoffs
Software encoders (libx264, libx265, SVT-AV1) produce the best quality per bit but are slow. Hardware encoders built into GPUs and CPUs encode 10–50x faster but produce larger files at equivalent quality — typically 20–40% larger for the same visual quality. Nvidia's NVENC supports H.264, H.265, and AV1 (RTX 40-series and later for AV1). AMD's AMF supports H.264 and H.265; AV1 support arrived with RDNA3. Intel Quick Sync supports all three on 12th-gen and later CPUs. The quality gap between hardware and software encoding has narrowed significantly. Nvidia's RTX 40-series NVENC AV1 implementation is competitive with SVT-AV1 at preset 6–8 for many types of content. For live streaming or real-time capture where software encoding isn't feasible, hardware AV1 is now a legitimate option. CocoConvert's conversion pipeline uses software encoding for all output formats. This is a deliberate choice: cloud conversion services need consistent, reproducible output quality across different server hardware, and software encoders provide that. The tradeoff is that very long files (feature-length movies, multi-hour recordings) take longer to process than they would on a local machine with a modern GPU. For a 2-hour 4K file converted to AV1, expect processing times in the 20–45 minute range depending on server load. We're transparent about this because it affects whether our service is the right tool for your workflow — if you're batch-converting a large video library, a local FFmpeg setup with hardware encoding may be more practical.
The Practical Bottom Line
The codec landscape in 2026 has a clearer hierarchy than it did three years ago. AV1 has won the web streaming battle — it's royalty-free, broadly supported, and delivers meaningfully better compression than H.265. H.265 remains the right choice for archival, professional workflows, and 4K HDR distribution where AV1 support is still inconsistent. H.264 is the compatibility baseline you fall back to when you genuinely don't know what will play the file. For most people converting video through CocoConvert, the decision tree is straightforward: choose 'Web Compatible (H.264)' for anything going to an audience you can't profile, 'Modern Web (AV1)' for your own website or app where you know your users have recent hardware, and 'Archive (H.265)' for footage you're storing long-term. What CocoConvert handles well: converting between these three codecs for standard web and archival use cases, applying sensible quality settings by default, and handling the container/metadata details (like that hvc1 tag) that trip up manual FFmpeg commands. What it doesn't replace: a local FFmpeg or Handbrake setup for batch-processing large libraries, professional mastering workflows that require frame-accurate editing metadata, or Dolby Vision HDR authoring. If you want to experiment with settings before committing to a full conversion, upload a short clip (30–60 seconds) and compare the outputs. File size and quality differences between codecs are immediately apparent on any actual content, and that hands-on comparison will tell you more than any benchmark chart.