H.265 vs H.264: What Actually Changes

They are successive generations of the same idea, published by the same ITU-T and ISO groups a decade apart.

The bigger blocks are most of the story. HEVC can describe a large flat area, like sky or a wall, as one unit instead of dozens, and it has more ways to predict a block from its neighbors and from other frames. The standard’s overview paper covers the full toolset. The 50% figure comes from the developers’ own subjective tests, published in IEEE TCSVT in December 2012, using reference encoders rather than the x264 and x265 builds people actually run.

It depends on the footage, far more than most comparisons admit. We encoded two clips with x264 and x265 at the same CRF and preset, video only, and scored every file with VMAF, where 100 means indistinguishable from the original.

On the busy clip at CRF 28, H.265 was smaller and scored higher. Against the H.264 file that came closest on quality, CRF 27 at VMAF 99.52, the H.265 file was 39.3% smaller. On the calm clip, the saving all but vanished.

Why so different? A slow pan across a still mountain scene is easy for both codecs. Most of each frame is predictable from the last one, and what is left is grain that neither can predict. There is little for HEVC’s extra tools to win. Fast motion and heavy detail are where the larger blocks and better prediction pay off. If your videos are mostly talking heads and screen recordings, expect the low end of that range. If they are handheld, sports or low-light, expect the high end.

In software, yes. x265 used 1.73 times the CPU time of x264 on the calm clip (28.4 against 16.4 CPU-seconds) and 1.76 times on the busy one (31.0 against 17.6). We report CPU time rather than wall-clock time because the test Mac was running other work. The ratio held on both clips.

In hardware the penalty disappears, and the comparison flips. Apple Silicon’s media engine encodes both codecs, and at the same 4,500 kbps target the hardware HEVC file scored VMAF 98.01 against 93.57 for hardware H.264. If you are using the hardware encoder for speed, pick HEVC.

# hardware encoders in ffmpeg; same bitrate, very different quality
ffmpeg -i in.mov -c:v hevc_videotoolbox -b:v 4500k -tag:v hvc1 out-hevc.mp4
ffmpeg -i in.mov -c:v h264_videotoolbox -b:v 4500k out-h264.mp4

Either hardware encoder is still far less efficient than software per byte. Software x265 reached VMAF 98.19 on the same clip in about a quarter of the hardware file’s size. Our 4K guide measures that trade-off in detail.

No, and this is the real cost. On caniuse’s data as of 27 September 2026, 97.26% of global browser usage fully supports H.264 video. For HEVC the figure is 16.94%, with another 76.59% marked partial. Chrome, Edge and Firefox all play HEVC only where the hardware and operating system provide a decoder, while Safari plays it everywhere.

On Windows, the decoder is a Microsoft Store item. The paid HEVC Video Extensions listing costs $0.99 and is rated 1.8 out of 5 from 3,725 ratings. A free build for devices whose manufacturer licensed the codec exists, but is not offered everywhere. It is the same gap that makes HEIC photos fail on Windows, because HEIC is an HEVC still.

On Apple devices it is a non-issue, with one trap. Apple says macOS High Sierra or later is needed to fully view and edit HEVC, and lists hardware HEVC decode among the M2 Pro’s media engine features. The trap is the tag: an HEVC MP4 that ffmpeg labels hev1, its default, was reported as not playable by AVFoundation in our test. The same encode labeled hvc1 played. Add -tag:v hvc1 to every HEVC export.

When you do not control playback. Email attachments, uploads to unknown systems, video on a public web page without an H.264 fallback. With H.264 at 97% full browser support against 17% for HEVC, that is not a close call. Our email guide recommends H.264 for exactly this reason.

When the footage is calm. A 3% saving is not worth any compatibility risk. Screen recordings, slideshows and locked-off interviews live at that end of the range.

When software encode time is the bottleneck. At 1.7 times the CPU cost, a big H.265 batch takes noticeably longer. Either accept it, or use the hardware HEVC encoder and accept larger files.

H.265 is the right choice for your own library on Apple devices, for 4K, and for busy footage you are storing rather than sending. That is where the 18 to 39% savings we measured on busy footage add up.

If you only need one codec switch, you do not need an app. ffmpeg and HandBrake both encode both codecs for free. Smol’s default video output is H.265 at 10-bit through libx265, and it switches to H.264 through libx264 in settings. Pinned to the same CRF and preset, its output was bit-identical to ffmpeg’s in our tests, down to matching MD5 hashes, so there is no quality reason to choose it.

If you want hardware encoding, Smol cannot do it. It has no VideoToolbox option. Use ffmpeg’s hevc_videotoolbox or HandBrake’s VideoToolbox encoders.

Smol is worth $29 if you convert video often enough that remembering -tag:v hvc1, CRF values and presets is a chore: drop a folder, get tagged, correctly encoded files, with the photos and PDFs handled in the same pass. The batch comparison shows what that does and does not change, and the format reference lists what it reads and writes. Our main video guide covers the free routes for a single file.