Skip to content

AV2

current

AV2 is the Alliance for Open Media’s next-generation video format and the successor to AV1. It is developed under the same royalty-free patent policy as AV1.

  • The specification is published. AV2 Bitstream & Decoding Process Specification v1.0.0 is dated 28 May 2026 and is the current release on av2.aomedia.org. AOMedia announced it in June 2026. Earlier working drafts, such as “v13” from January 2026, are superseded; despite the numbering they are older than 1.0.0.
  • The coding tools are frozen. AOMedia says future versions may add profiles and metadata but the core coding tools will not change.
  • Only reference software exists. AVM v1.0.0 is the reference encoder and decoder. VideoLAN’s dav2d decoder is in early development (0.0.x tags). There is no production encoder, no FFmpeg support, no browser support and no hardware decoder.
  • No container mapping yet. There is no published AV2 binding for MP4/ISOBMFF or Matroska. Do not reuse AV1’s av01/V_AV1 identifiers for AV2; keep test output as raw elementary streams from AVM.

AOMedia quotes about 30% better compression than AV1, measured with VMAF and PSNR using the AVM reference encoder under the working group’s test conditions. That compares reference encoders at very slow settings; it says nothing yet about how production AV2 encoders will compare with SVT-AV1 or aomenc at practical speeds.

AV2 is therefore for research, evaluation and implementers. Don’t plan delivery around it until encoders, decoders and container bindings ship.

AV2 keeps AV1’s overall design (OBUs, superblocks, multi-symbol entropy coding, CDEF, loop restoration, film grain synthesis) and adds or replaces many tools. Some of the larger ones named in the specification:

  • superblocks up to 256×256;
  • optical-flow refinement of motion vectors;
  • trellis-coded quantisation;
  • a cross-component sample offset (CCSO) filter, plus cross-component prediction and transforms for chroma;
  • extended screen-content tools (palette and intra block copy);
  • multi-layer and multi-stream coding: scalable layers, stereoscopic and multi-view video, and several streams in one bitstream.

Version 1.0.0 defines one toolset (Main) and these profiles:

ProfileBit depthChromaInteroperability point
Main_420_10_IP08, 104:0:0, 4:2:00
Main_420_10_IP18, 104:0:0, 4:2:01
Main_420_10_IP28, 104:0:0, 4:2:02
Main_422_10_IP18, 104:0:0, 4:2:0, 4:2:21
Main_444_10_IP18, 104:0:0, 4:2:0, 4:4:41
ConfigurableSignalledAny of the aboveSignalled

All v1.0.0 profiles stop at 10-bit. AOMedia has started work on a 12-bit profile for professional and cinema use. The interoperability point (IP) sets how many layers a decoder must handle at once: IP0 allows up to four streams with one layer each, IP1 allows up to two embedded (scalable) layers, and IP2 up to three, with streams and layers combined. It is not a speed or quality setting. Configurable is for streams that fit none of the named profiles; their limits come from the sequence header.

Levels run from 2.0 to 8.3, with a High tier available from level 4.0. As with AV1, they limit picture size, decode rate and bitrate, not quality.

Build AVM at the v1.0.0 tag rather than the development branch, whose bitstream can differ from the published specification. AVM’s encoder is a research encoder: it is very slow, and its results describe AVM at that tag, not what production AV2 encoders will eventually achieve. Decode with the AVM decoder from the same tag.

AV2 implementations
RelationshipProject or formatEvidence and scopeStatus
Implements byAVM v1.0.0AVM v1.0.0 is the reference software accompanying AV2 v1.0.0.Scope: Reference Software · Sources: AV2 Bitstream & Decoding Process Specification v1.0.0Normative

For browser and device status as it changes, see video codec support.