SVT-AV1 presets compared

SVT-AV1’s preset dial runs from 0 to 13 and the guidance is usually “lower is better, pick what you can afford”. One rung of it turns out to be worth nothing at all.

Short version. Preset 12 was faster AND smaller than preset 10. If you are on 10, moving to 12 costs you nothing and gains you both.

Variant	Encode	Size	Cost
preset 4	2399 ms	174.6 KB	$0.0058
preset 6	1585 ms	197.3 KB	$0.0043
preset 8	1105 ms	224.4 KB	$0.0028
preset 10	653 ms	229.7 KB	$0.0021
preset 12	614 ms fastest	219.7 KB	$0.0021

Measured 2026-08-20 on the Rendobar API. Encode time is the FFmpeg step alone, separated from download and upload. 5 of 5 runs succeeded. Sizes are exact and repeatable. Timings are a single sample and vary up to 2x run to run, so treat small differences as noise. Why.

Preset 10 is dominated#

Preset 12 finished 39 ms sooner and produced a file 10 KB smaller. There is no axis on which preset 10 wins, which makes it a rung with no reason to exist for this content.

This is the same non-monotonicity we found in FFmpeg x264 presets compared, and it has the same cause. -crf targets constant quality, not constant size. A preset that searches harder can find detail worth spending bits on, so output size at fixed quality is an emergent property rather than a dial you are turning.

The lesson generalises past AV1: if you are choosing an encoder preset by output size, measure it. Do not assume the ordering.

The dial is real, and shorter than advertised#

Preset 4 to preset 12 is 3.91x in time and 25.8% in size.

That 3.91x is worth having. What is more surprising is the size column: the slowest setting measured bought only about a quarter fewer bytes than the fastest. AV1’s reputation is that the slow presets are transformative, and on this clip the whole usable range is 174.6 KB to 229.7 KB.

For most pipelines that makes preset 8 the sensible default. It is 2.2x faster than preset 4 for 28% more bytes, and it was also the setting that beat VP9 outright in AV1 vs VP9 vs HEVC vs H.264.

Go slower than 8 when you encode once and serve many times, and the 25% matters more than the four seconds. Go faster when you are transcoding a library that is mostly never watched.

These timings are trustworthy, and that is not a given#

Three runs per preset, and the spreads were 1.3% to 3.4%.

That is worth flagging because it is unusual on this infrastructure. The same libx264 command on default threading varied by 85% across five runs, which was enough to retract a claim we had already published.

SVT-AV1 does not behave that way. It manages its own thread pool and pins its own work, so it does the same amount of computation regardless of what else the machine is doing. The result is a benchmark you can read at single-digit percentage resolution without pinning anything yourself.

Sizes were byte-identical across all three runs of every preset, as they are everywhere on this site.

Where this stops#

One 5-second 1280x720 clip, libsvtav1 -crf 35, audio stripped, three runs per preset. Presets 0 to 3 were not measured: they are slow enough that the sweep would have dominated the cost of every other benchmark here, and almost nobody runs them outside archival work.

Content matters for the non-monotonicity specifically. The preset 10 versus 12 inversion is a property of what those search strategies found in this footage, and it may not reproduce on yours. The 3.91x speed range and the rough size envelope should travel.

AV1 is one of several codecs measured on this source. The comparison against H.264, HEVC and VP9 is in Video codec comparison, measured.