The AV2 Codec Is Here: What the 1.0.0 Release Means for Open Video (And Why You Should Care)

 

AV2 1.0.0 is here: royalty-free, 20–30% better than AV1, with dav2d decoder. What it means for video encoding & when to adopt.

The Video Codex You’ve Been Waiting For (Without the Royalties)

If you’ve ever compressed a video for the web, you know the trade-off all too well: file size versus quality. H.264 is aging. HEVC (H.265) works well but comes with a labyrinth of patent pools and licensing fees. AV1 solved many of those problems—but it arrived with high encoding complexity and slow adoption cycles.

Now imagine a codec that gives you even better compression than AV1, remains completely royalty-free, and is backed by the same open-source alliance that brought you AV1. That codec is now a reality.

The Alliance for Open Media (AOM) has officially tagged version 1.0.0 of its reference software for AV2, named AVM (AOM Video Model). 

Alongside this, VideoLAN has published dav2d—an open-source AV2 decoder built in the spirit of the wildly successful dav1d. After years of drafts and incremental milestones, AV2 is no longer a future promise. It is a present-day specification.

This post walks you through what AV2 is, how it improves on AV1, what the 1.0.0 release actually means, and what you should (and shouldn’t) do with this information today.

What You Will Learn

• The technical lineage from AV1 to AV2 and why a “clean-slate” design matters.

• Key compression improvements in AV2 (block partitioning, transforms, and in-loop filtering).

• Why the dav2d decoder is as important as the spec itself.

• Realistic adoption timelines for web browsers, hardware, and streaming platforms.

• Common mistakes developers make when evaluating a new codec (and how to avoid them).

From AV1 to AV2: A Brief History of Open Video

To understand AV2, you need to understand what AV1 achieved—and where it fell short.

AV1 was released in 2018 as a response to the fragmented, patent-encumbered landscape of HEVC. Backed by AOM members including Google, Mozilla, Cisco, Netflix, and Amazon, AV1 proved that a modern, royalty-free codec could compete with—and often beat—proprietary alternatives. 

Today, AV1 is supported by every major browser, most streaming platforms (YouTube, Netflix, Vimeo), and a growing range of hardware decoders (Intel, AMD, NVIDIA, Apple).

But AV1 was never meant to be the final word. Its design was shaped by compromises needed to get a working standard out the door quickly. Encoding remained notoriously slow (despite optimizations in libaom and svt-av1), and some of its tools were inherited from VP9 and earlier prototypes.

AV2 was conceived as the next step: a more aggressive, clean-slate design that could push compression efficiency further while maintaining the same royalty-free mandate. Work on AV2 began in earnest around 2021, with drafts circulating since early 2023. 

The release of AVM 1.0.0 on May 29, 2026 marks the first time the reference implementation has been declared stable and feature-complete.

Why “reference software” matters

AVM is not production-ready encoder software. It’s the “model” that defines exactly how AV2 works—every block partition, every transform, every bitstream decision. Hardware designers, encoder developers, and decoder implementers use AVM to validate their own implementations. 

When AVM hits 1.0.0, it signals that the specification is frozen and ready for real-world deployment.

What’s Actually New in AV2? (Beyond “Better Compression”)

Better compression efficiency is the headline, but the how matters for anyone who builds or uses video tools. Here are three core areas where AV2 departs from AV1.

Smarter Block Partitioning (CDT and MTT)

AV1 used a recursive block partitioning scheme with 4-way splits (horizontal, vertical, and their combinations). AV2 introduces Conditional Directional Transforms (CDT) combined with a more flexible Multi-Type Tree (MTT). 

• In plain English: AV2 can divide a frame into shapes that better match real-world image edges and textures, reducing the number of “residual” blocks that need to be encoded.

• Practical impact: Less bitrate spent on block artifacts, especially in high-motion scenes or areas with diagonal lines.

 Improved In-Loop Filters (CNR and CDEF-2)

Block-based codecs always produce visible boundaries between blocks. AV1 had a deblocking filter plus a constrained directional enhancement filter (CDEF). AV2 adds a Cross-Component Non-local Reconstruction (CNR) filter that uses information from multiple color planes (Y, U, V) to smooth artifacts more intelligently. A

dditionally, CDEF has been upgraded to CDEF-2, which adapts more aggressively to local texture.

Practical impact: Less “banding” in gradients, cleaner edges, and better performance on noisy or high-frequency content.

Larger Transform Sizes and Learned Transforms

AV2 supports transform blocks up to 256×256 (AV1 maxed at 64×64). This allows more efficient encoding of large smooth areas (like skies or walls) without chopping them into tiny pieces. 

Even more interesting: AV2 includes learned transforms—pre-trained matrices that replace some of the fixed discrete cosine (DCT) and sine (DST) transforms. These learned transforms are static (not per-video), so no extra overhead, but they better match statistical properties of natural images.

Practical impact: Significant gains on high-resolution content (4K, 8K) and computer-generated imagery (screen recordings, UI captures).

Table: AV1 vs. AV2 at a Glance

Estimated compression gain over AV1 (based on AOM internal tests and early third-party benchmarks): 20–30% lower bitrate for same visual quality, with larger gains on high resolution and synthetic content.

Why dav2d Is as Important as the Spec

A codec spec is useless without a decoder. And not just any decoder—a fast, portable, open-source decoder that can run on everything from a $10 IoT device to a flagship smartphone.

VideoLAN’s dav1d became the gold standard for AV1 decoding because it outperformed the reference decoder (libaom) by a wide margin, especially on lower-end hardware. dav2d is built on the same principles:

• Hand-optimized assembly for ARM (NEON) and x86 (AVX-512, AVX2).

• Frame-level parallelism to leverage multi-core CPUs.

• Low memory footprint suitable for embedded devices.

With dav2d already published as open source (the initial commit dropped on May 15, 2026), browser teams (Mozilla, Google, Apple) can begin integrating it into Firefox, Chrome, and Safari. 

The timeline for native browser support is typically 12–18 months after a stable decoder becomes available, but we may see experimental flags sooner.

Key insight: The release of dav2d before the final spec was even frozen was a strategic move. It forces the ecosystem to think about real-world decoding performance from day one, not just spec compliance.

Practical Examples: Where AV2 Will Shine First

Example 1: Streaming services (Netflix, YouTube, Vimeo)

These platforms have already invested heavily in AV1 encoding pipelines. Switching to AV2 will require re-encoding their entire catalog (a massive compute cost), but the storage and bandwidth savings are compelling. 

For a service like Netflix, a 25% reduction in bitrate at the same quality could save petabytes of bandwidth per month. Expect them to start with new, high-demand content, then backfill over years.

Example 2: Video conferencing (Zoom, Google Meet, Teams)

Low-latency, real-time encoding is hard. AV1’s encoding complexity made it a poor fit for software-based video calls on low-power devices. AV2’s improved block partitioning and transforms actually reduce decision complexity in some scenarios (though encoding remains heavy). 

The bigger win is decoding efficiency: dav2d on a laptop’s battery can decode 1080p AV2 at lower power than AV1, thanks to the new filters and transform optimizations.

Example 3: User-generated content (discord, Twitch clips, mobile uploads)

For individual creators, AV2 won’t matter until hardware encoders appear. But once they do (think: Qualcomm, MediaTek, Apple silicon in 2028–2029), mobile users will upload videos at half the file size for the same quality. That means faster uploads, less mobile data usage, and cheaper cloud storage for platforms.

Common Mistakes to Avoid (When Evaluating a New Codec)

Mistake 1: Benchmarking with the reference software

AVM is correct, not fast. If you run AVM against libaom or x265 and declare “AV2 is too slow,” you’re missing the point. The reference implementation is for validation, not production. Real-world encoders (like a future svt-av2) will be 10–100x faster.

Mistake 2: Assuming immediate browser support

Even with dav2d ready, browser integration requires consensus on security, sandboxing, and performance. Safari, in particular, has been slow to adopt AV1. AV2 will follow a similar curve: 6–12 months for experimental flags, 2–3 years for default-on support across all major browsers.

Mistake 3: Neglecting the hardware decoding gap

Software decoding of AV2 at 4K/60 on a laptop is fine. On a smartwatch or a budget Android phone, it’s not. AV2 will only replace H.264/HEVC/AV1 at scale when dedicated decoder blocks ship in SoCs. That typically starts 2 years after a spec freezes. So look for 2028 devices to include AV2 hardware decoders.

Mistake 4: Re-encoding everything right now

Don’t do it. Without wide playback support, serving AV2 will force fallbacks to AV1 or H.264, negating any bandwidth savings. The smart strategy: encode to AV2 as a master archive format (future-proofing), but serve AV1/H.264 to users until AV2 adoption crosses 80% (likely 2029+).