AV1 is the royalty-free video codec the Alliance for Open Media (AOMedia) shipped to break the patent-licensing logjam around HEVC. It's been a standard since 2018 and at this point it's not theoretical anymore — Netflix, YouTube, Meta, and Twitch all serve AV1 to clients that can decode it. But "serving AV1 to clients that can decode it" hides most of the operational decisions, and "production AV1" is a different conversation than "AV1 the codec."
This page is the engineering reference: what AV1 actually is, where it wins, where it loses, and what owning AV1 in production looks like as of 2026.
The short version
AV1 gives you roughly 30% better compression than HEVC at the same visual quality, and roughly 50% better than H.264. That number comes with caveats — it's measured on Netflix's content set, with PSNR/VMAF as the proxy, with libaom or SVT-AV1 at high CPU budgets. Your numbers will be lower if your content is sports or live, higher if your content is animation. But the directional answer is: AV1 is a real codec gain, not a marketing one.
The cost is encoding compute. libaom-av1 in 2018 was unusably slow. SVT-AV1 in 2026 is realistic — preset 6-8 is in the same order of magnitude as x265 medium, with quality that's competitive — but it's still 2-4x slower than the H.264 baseline most pipelines were sized for. Hardware encoding helps if you have it. Decode side, AV1 is now in every recent flagship phone, every Chromium-based browser, and most smart TVs from 2022 forward — practically, though, you still need an HEVC fallback for the long tail of devices.
What AOMedia got right
AOMedia is the industry's response to the HEVC patent-licensing fragmentation. Three patent pools (MPEG LA, HEVC Advance, Velos Media) plus uncategorized patent holders meant nobody could write an HEVC encoder or decoder without a high-friction conversation with lawyers. AV1's promise: royalty-free, no patent pool, no per-stream license.
That promise has held up. Sisvel filed an AV1 patent pool announcement in 2020 that AOMedia rejected as an attempt to encumber a clean codec; Sisvel has not gotten meaningful traction. The codec is shipping royalty-free in practice, and AOMedia maintains a defense fund for members.
Technically, AV1 borrows from VP9 and adds aggressive tools:
- Super-resolution — encode at lower resolution, signal upscale to the decoder. Effective when bandwidth-starved.
- Film grain synthesis — rather than encoding film grain as noise (which compresses badly), strip it and send a synthesis seed. The decoder regenerates noise that looks right. Real bitrate savings on grainy content.
- Adaptive transform sizes — 4×4 up to 64×64. The encoder picks per-block.
- Loop restoration filters — Wiener and self-guided projection filters reduce blocking artifacts at very low bitrates.
- OBMC and warped motion — subtler motion compensation than HEVC's translational model. Helps on rotation/zoom content.
The standard itself is stable; the encoder ecosystem is what's evolved.
Encoder economics
The codec is the spec. The encoder is the production reality. There are three encoders worth knowing:
libaom-av1 (reference)
The AOMedia reference encoder. Highest quality at any preset, slowest by a margin. cpu-used 0 is research-grade — minutes per frame. cpu-used 6-8 is what production used to live at; even there, it's slower than x265 placebo. libaom is what the academic VMAF numbers come from; it's not what most of production runs.
SVT-AV1 (Intel/Netflix)
Scalable Video Technology AV1. Built for parallelism — wider thread scaling than libaom, lower per-thread efficiency. The mainline production encoder in 2026. Presets 0-13; production VOD lives at preset 6-8, live encoding at preset 10-12. Quality at preset 6 is within a few percent of libaom cpu-used 4, at a fraction of the wall time. Netflix runs SVT-AV1 for their AV1 ladder.
rav1e (Mozilla)
Rust-implemented AV1 encoder. Slow, but the safety story and the build hygiene make it appealing for hostile-input pipelines. Not a frontline production choice for compression efficiency, but worth knowing about for sandboxing-conscious deployments.
The practical decision in 2026: SVT-AV1 for most VOD, libaom-av1 if you're willing to pay for top-end quality on premium content, and you fall back to your existing H.264/HEVC ladder for everything else.
Hardware encoding
NVIDIA Ada (RTX 40-series, 2022+) and Intel Arc were the first consumer cards to ship AV1 encoders. Datacenter-grade AV1 encoding hardware (Google's VCU, NETINT VPUs, AMD MA35D) has been available since 2023 and is what makes live AV1 economically real. Without hardware AV1, live encoding is a CPU problem you don't want.
NVENC AV1 quality has gotten close to SVT-AV1 preset 8 — not as good as preset 4-6, but acceptable for live. The cost-per-stream calculation flips when you can amortize a $5K accelerator across 50+ AV1 streams.
For VOD, hardware AV1 buys you throughput, not better compression. SVT-AV1 software encoding still wins on quality at the same bitrate; hardware wins on $/encode.
Hardware decoding
This is the practical adoption gate. AV1 software decode is fine on modern x86 — dav1d (the VLC/FFmpeg AV1 decoder) is fast enough to play 4K AV1 on any 2020+ desktop. The problem is mobile: software AV1 decoding crushes battery on phones.
Hardware AV1 decode adoption as of 2026:
- Mobile — every Android flagship since Snapdragon 8 Gen 1 (2022), every iPhone since 15 Pro (2023). Mid-range Android lags by 1-2 generations. Practical install base for "AV1-capable mobile": ~70% of new traffic, less than 50% of total install base.
- Desktop — every recent Intel (12th gen+), AMD (RDNA2+), NVIDIA (RTX 30+) GPU. Browsers expose this via VideoDecoder API.
- TV — Samsung 2020+, LG 2022+, Sony 2022+, Roku Ultra 2022+. The long tail of 2018-2020 smart TVs is HEVC at best.
Conclusion: AV1 is shipping to a meaningful slice of your audience, but you cannot drop your HEVC ladder yet. Three-codec ladders (AV1 → HEVC → H.264) are the production reality for 2026 and will be for at least 2-3 more years.
When AV1 wins
- Premium VOD — Netflix, YouTube, Disney+ premium tiers. Compression efficiency at the same bitrate translates directly into bandwidth bill reduction. The encoding cost is a one-time hit; the bandwidth saving compounds per stream.
- Mobile-heavy distribution — capable mobile GPUs decode AV1 efficiently. Bandwidth on mobile networks is more expensive than CPU.
- Animation / low-noise content — AV1's transform tools and film-grain synthesis don't help here, but the residual coding gains compound on flat, predictable content. Animation studios see VMAF wins of 2-3 points over HEVC.
- UGC platforms — YouTube, TikTok-class scale. Even small bitrate savings × billions of streams = real money. UGC is where AV1 dollar value is highest.
When AV1 loses
- Live, low-latency — encoding latency at high quality is a problem. Hardware AV1 helps, but live AV1 is still a "talk to your hardware vendor" conversation, not a default.
- Long-tail device support — anything before 2020 is HEVC at best. If your audience skews older or developing-market, AV1 isn't reachable.
- High-grain content — sports under stadium lights, low-light interview footage. AV1 film-grain synthesis helps with film grain but not sensor noise. HEVC's larger codebook handles noisy content with less artifact.
- Encoding budget constraints — if you're at "I need to encode this in real-time on the same machine I'm running Postgres on," AV1 isn't your codec yet.
How to think about your AV1 ladder
The pragmatic 2026 ladder for a mid-size streaming platform:
- AV1 top tiers — 4K, 1440p, maybe 1080p. SVT-AV1 preset 6-8. Serve to AV1-capable clients (Chromium browsers, modern Android, recent iOS, smart TVs from 2022+).
- HEVC mid tiers — 1080p, 720p. x265 medium. Serve to non-AV1 capable Apple/Safari (FairPlay-enabled), older Android, mid-range smart TVs.
- H.264 floor — 720p down to 240p. x264 medium. Serve to anything that doesn't decode the others, including downlevel browsers and developing-market devices.
Bitrate savings of AV1 over HEVC are real but rounding-error compared to the cost of dropping HEVC entirely. Three-codec is the production answer until ~2028.
A note on AV1 vs VVC
VVC (H.266, MPEG's successor to HEVC) is the only other 2026-era codec in AV1's compression-efficiency tier, and the comparison comes up in every codec-roadmap conversation. VVC has slightly better compression than AV1 in independent testing — single-digit percentage points on most content. But the patent-licensing story around VVC is back in three-pool territory (MPEG LA, Access Advance, plus VVC's own complications), and hardware decode in shipping consumer devices is essentially zero as of mid-2026. AV1's encoder ecosystem and hardware-decode install base put it years ahead of VVC for production deployment, regardless of the codec-spec comparison. If you're building a 2026 ladder, AV1 is the answer to the codec question and VVC is the watch-this-space answer to the same question for 2028 onward.
What MpegFlow does with AV1
MpegFlow's DAG runtime runs SVT-AV1 on FfmpegExecutor workers with per-tier preset budgets configured in workflow YAML. The partitioner persists each rendition stage to job_stages with explicit dependency tracking, so AV1 top tiers and HEVC mid tiers run as parallel sibling stages on appropriate worker pools (GPU pool for NVENC HEVC, CPU pool for SVT-AV1 software). Per-stage retry handles transient failures; sibling cancellation propagates fatal failures across rendition stages; rendition-level partial-success reporting surfaces granular per-stage state when a slow AV1 rung fails and the cheaper rungs succeeded.
The KEDA-driven autoscaler sizes the worker pools independently to the queued workload, so AV1 top-tier work doesn't bottleneck H.264/HEVC rungs queued behind it.
For live, hardware AV1 (NVENC AV1) is configurable as a per-rendition encoder choice when the GPU pool is available; most live customers ship HEVC live and AV1 VOD because the operational economics still favor that split for now.
If you're evaluating an AV1 rollout and want a sounding board on ladder design, encoder choice, or the device-mix question — we run that conversation regularly with design partners.