The 2026 Standards Baseline
This sheet fixes the reference points the rest of the portal compares against. It is a statement of observed practice, not a preference. Where an industry convention has settled, we record it; where it is still moving, we say so and show the spread.
Three codecs, three jobs
H.264 remains the widest decoder base ever deployed. Its encoding expense is low and its tooling is mature, so it still anchors the low and mid rungs of most ladders. HEVC spends more compute per frame to reach the same distortion at a lower bitrate, which pays off at high resolution. AV1 pushes further on prediction and entropy coding and earns its keep on the top rungs, where a percent of bitrate is worth real money across a large audience.
| Codec | Typical rung | Quality mode | Decoder base |
|---|---|---|---|
| H.264 | 360p-1080p | rate factor 20-24 | Very wide |
| HEVC | 1080p-2160p | rate factor 22-28 | Wide at 10-bit |
| AV1 | 1440p-2160p | rate factor 28-36 | Growing, uneven |
The numbers above are ranges, and ranges are the honest unit. A single figure would hide the influence of grain, motion and shot length. A two-hour concert film and a ninety-minute talk recording do not share one rate factor, even at the same resolution.
How the lab states a baseline
Each accepted baseline names the source, the colour profile, the frame rate, the quality mode, the keyframe interval and the reconstruction depth. Two encoders are only compared when all six agree, or when a difference is itself the subject of the run. This rule removes most of the noise that makes online benchmark tables contradict each other.
What shifted during the year
Two changes stand out. First, ten-bit reconstruction moved from a niche setting to a default for high-definition tiers, because it removes banding on gradients with almost no bitrate penalty. Second, constant-quality rate control became the normal frame for archival and on-demand work, with two-pass bitrate targeting kept for live and fixed-channel cases.
The baseline is revised when a shift is reproduced on the bench by two independent operators. Speculative release notes are not enough. Until a change survives that test, it stays in the an open question rather than a row in the table.
Reading the baseline against your own source
Treat Table 1 as a starting bracket, then move the rate factor until the distortion curve flattens. The companion pages on constant quality and measurement explain how to read that curve. The protocol explains how to record the run so the result can be checked by someone else.