Updated November 2026: the VP8 vs VP9 short answer now opens the article, libvpx v1.17.0 (August 2026), FFmpeg 8.1 and two 2026 libvpx security fixes added, new FAQ on the VP9 release date, the vp09 codec string and VP9 live streaming.
VP8 vs VP9: The Short Answer
VP8 is Google's royalty-free video codec from 2010. It compresses at roughly H.264 Baseline level, encodes fast, and is one of the two video codecs every WebRTC browser must implement under RFC 7742. VP9 is its 2013 successor, with about 40–45% lower bitrate than VP8 at the same quality in independent tests, HEVC-class compression, 10- and 12-bit HDR profiles, and 3–8× the encoder CPU of VP8 in libvpx.79
Use VP8 when every WebRTC peer must be able to decode the stream and encoder CPU is tight. Use VP9 for royalty-free VOD and 4K renditions, for SVC conferencing between peers that support it, and as the rung for devices that decode VP9 but not AV1. New streaming ladders in 2026 start from AV1 and keep VP9 as the fallback.
| Criterion | VP8 | VP9 |
|---|---|---|
| Released as open source | May 2010 | June 2013 |
| Block structure | fixed 16×16 macroblocks | 64×64 superblocks, quadtree down to 4×4, rectangular splits |
| Motion vectors | 1/4 pixel, 6-tap filter | 1/8 pixel, 8-tap filter |
| Transforms | 4×4 only | 4×4 to 32×32 DCT plus ADST |
| Bitrate at equal quality | reference | about 40–45% lower than VP8 |
| Bit depth and HDR | 8-bit 4:2:0 only | Profiles 0–3: 8/10/12-bit, 4:2:0 to 4:4:4; HDR via Profile 2 |
| Scalability for SFUs | temporal layers | temporal and spatial layers (SVC) |
| Encoder CPU (libvpx) | real time on almost any device | 3–8× VP8 at comparable settings |
| WebRTC status | mandatory to implement (RFC 7742) | optional, negotiated in SDP |
| Containers | WebM, Matroska, RTP | WebM, MP4, Matroska, RTP |
| Where it fits in 2026 | WebRTC floor codec, legacy WebM | VOD and 4K fallback under AV1, SVC calls |
Table 0. VP8 vs VP9 at a glance. The sections below unpack each row.
VP9 vs H.265 (HEVC). VP9 trails HEVC by about 5–15% in bitrate at matched objective quality, and the gap narrows on low-resolution, low-bitrate content.7 VP9 wins on licensing (an irrevocable Google patent grant and no content royalties) and on browser reach: Chrome, Firefox, Edge and Safari all decode it, while HEVC playback in the browser still depends on a hardware decoder. HEVC wins on hardware encode in cameras, phones and Apple devices. For VP9 vs H.264 and the other head-to-head pairs, see our video codec comparison table.
Why This Matters
If you build a real-time video product (video calling, conferencing, customer-support video, live shopping, online tutoring, telemedicine), you already ship VP8, planned or not, because every WebRTC browser is required to implement it. If you build a streaming product, you have to decide whether to encode a VP9 ladder for the long tail of devices that decode VP9 but not AV1 (every Android phone since 2016, every Chrome and Firefox install, Safari on iOS 14 and later, most smart-TV chipsets since 2015). And if you work anywhere in the open-source video pipeline, you almost certainly link against libvpx, the reference implementation Google maintains for both codecs and the encoder FFmpeg calls for VP8 and VP9 (FFmpeg decodes both with its own native decoders by default).
This article explains what VP8 and VP9 are and traces the On2 → Google → AOMedia lineage that produced them. It then lays out the technical differences from the H.264 / H.265 family and closes with the 2026 deployment question every video product team runs into: where each codec is still worth shipping, and where AV1 has already taken over. You do not need prior codec knowledge to follow it, since we define every term in plain language before using it. If you have not yet read H.264 / AVC: The Workhorse Of The Internet and H.265 / HEVC: +50% Over H.264 And The Patent Nightmare, start there: VP8 was designed as a royalty-free counterpart to H.264, and VP9 as a royalty-free counterpart to HEVC.
What VP8 And VP9 Actually Are
A codec is two pieces of software glued together. An encoder takes raw pictures from a camera or a video editor and packs them into a small file, and a decoder reverses the process to put the pictures back on screen. The name is a portmanteau of coder + decoder. VP8 and VP9 are codecs in this sense, just like H.264 and H.265, and they slot into the same place in a video pipeline.
The naming predates Google's involvement. The VPx family was developed by a small US company called On2 Technologies (founded in 1992 as The Duck Corporation, a name that still shows up in some Matroska container codec tags). On2 shipped VP3 in 2001 (later donated to the Xiph.Org Foundation and renamed Theora), followed by VP4, VP5, VP6, and VP7, a run of proprietary codecs sold mostly to Adobe Flash for early online video. VP8 was On2's last release as an independent company, announced on 13 September 2008.1
The big event for VP8 came eighteen months later. On 5 August 2009 Google announced it was acquiring On2, and on 19 February 2010 the deal closed for $124.6 million.3 Three months later, on 19 May 2010 at Google I/O, Google did something nobody in video had done at this scale. Instead of monetising On2's patents, it released the VP8 specification under an irrevocable royalty-free patent licence, open-sourced the reference encoder and decoder as libvpx under a BSD-style licence, and bundled VP8 with the Vorbis audio codec inside a Matroska-derived container called WebM.1 The goal was stated openly: give the web a royalty-free alternative to H.264, whose MPEG LA patent pool had not yet settled whether it would charge per-stream royalties on internet video.
VP9 followed the same model. Development began in late 2011, and the first VP9-encoded bitstreams went live on YouTube in June 2013, the same month the spec was frozen.2 The compression target was simple: halve the bitrate of VP8 at the same perceptual quality, which would put VP9 in roughly the same position against HEVC that VP8 held against H.264. VP9 was royalty-free from day one, again under an irrevocable Google patent grant and again with libvpx as the reference implementation.
The third chapter is AV1. It is not a VPx codec, but it descends directly from that lineage. In September 2015, six months after the HEVC Advance patent pool launched and made HEVC licensing look unworkable, Google, Amazon, Cisco, Intel, Microsoft, Mozilla, and Netflix founded the Alliance for Open Media (AOMedia). Its first project was a royalty-free successor codec built by merging the best ideas from Google's VP10 (the in-progress VPx successor), Cisco's Thor, and Mozilla's Daala. The result was AV1, published in March 2018 and designed to retire VP9 the way VP9 had been designed to retire VP8. We cover AV1 in detail in AV1: The New Internet Standard And Where It Stands In 2026.
That is why this article covers VP8 and VP9 together and leaves AV1 out. The two VPx codecs share a corporate parent, a licence model, a reference implementation, and a container. AV1 inherits the licence model and the reference-implementation pattern, but AOMedia rebuilt its bitstream from scratch.
Figure 1. The VP8 → VP9 → AV1 timeline. Each generation took roughly five years, and each step moved to a new corporate vehicle (On2 → Google → AOMedia).
A Short History: From Flash Plug-in To WebRTC Default
VP8 and VP9 exist to give the web a video codec that nobody has to pay royalties to use. That sounds obvious in 2026, but in 2008–2010 nobody knew whether browser-delivered video would be subject to per-stream royalties, and Google's $124.6M purchase of On2 was a direct response to that uncertainty.
The chain of decisions started outside Google. The HTML5 <video> element was standardised at the W3C between 2007 and 2010 without a mandatory codec, because the working group could not agree on one. Apple wanted H.264, which it already supported across iOS and Safari. Mozilla and Opera wanted Theora (the open codec descended from On2 VP3) because they refused to ship a patent-encumbered codec inside a free browser. Microsoft wanted H.264. The video industry, which relied on Adobe Flash for cross-browser playback, followed the dispute closely.
Google's release of VP8 in May 2010 changed the discussion. VP8 was technically close to H.264 baseline profile, with roughly the same compression efficiency on simple content and somewhat worse results on hard content, but it was real, it was shipping, and it came with an irrevocable royalty-free patent grant. Mozilla and Opera shipped VP8 within months. Adobe added VP8 to Flash. Google added VP8 to Chrome and to YouTube. Microsoft and Apple did not adopt VP8 directly, but the existence of a royalty-free option pushed MPEG LA in August 2010 to commit publicly that H.264 would never charge for free-to-end-user internet video. That removed the immediate threat that had started the codec war.1
The second turning point was the WebRTC standard. When the IETF and W3C codified browser-to-browser real-time video between 2011 and 2016, the working group again could not agree on a single mandatory codec. The compromise, published as RFC 7742 in March 2016, made both VP8 and H.264 Constrained Baseline mandatory-to-implement in every WebRTC browser.4 That decision is why VP8 still has 100% browser support in 2026, fifteen years after release. Every Chrome, Firefox, Safari, and Edge install on every desktop and every Android phone ships a VP8 encoder and decoder because the WebRTC spec requires it.
The third turning point was YouTube's adoption of VP9 between 2013 and 2018. YouTube moved its 1080p and 4K streaming tiers to VP9 to avoid per-stream royalty exposure on H.264 and HEVC at its scale. As a result, Android device makers, smart TV vendors, set-top box makers, and browsers all had to support VP9 hardware decode or lose access to YouTube's catalogue. By 2018, VP9 hardware decode was effectively universal on Android phones manufactured after 2016 and on smart TVs manufactured after 2015.2
The fourth and last turning point was AV1. Once AOMedia shipped AV1 1.0 in 2018 and its members announced they would replace HEVC and VP9 with AV1 wherever they could, greenfield deployments started asking whether to "skip VP9 and go straight to AV1". YouTube began serving AV1 in 2018, and Netflix, which added AV1 in 2020, reported in December 2025 that AV1 powers about 30% of its streaming, with H.264 still its most-used codec.6 VP9 stayed in use, but its role shifted from next-generation default to the H.264 alternative you ship for the device tail too old to decode AV1.
That history sits behind every 2026 video product decision involving VP8 or VP9. Neither codec has a marketing budget, a patent-pool fight, or a new feature roadmap. They are infrastructure: VP8 is the WebRTC video codec, VP9 is the open-source middle codec between H.264 and AV1, and libvpx keeps both current because the video industry depends on them whether or not anyone plans to deploy them next.
How VP8 Works, In Plain Language
VP8 is a hybrid block-based codec with the same broad architecture as H.264. We unpack the hybrid framework in Hybrid Video Codec Architecture. The key idea here: VP8 cuts each frame into small rectangular blocks, predicts each block from its neighbours, encodes only the small leftover (the residual), and uses an entropy coder to write the residual bits efficiently into the bitstream.
VP8 differs from H.264 in four concrete design choices.
Block size. VP8 uses fixed 16×16 macroblocks like H.264 did, with sub-partitions down to 4×4 pixels for intra-prediction and 4×4 for motion compensation. There is no quadtree split.
Intra-prediction modes. VP8 offers 4 intra-prediction modes for 16×16 luma blocks (DC, vertical, horizontal, TrueMotion) and 10 for 4×4 blocks (adding six diagonal directions). H.264 has 9 for 4×4 blocks and 4 for 16×16. The numerical gap is small.
Motion compensation. VP8 supports quarter-pixel motion vectors with a 6-tap interpolation filter (compared to H.264's 6-tap filter at half-pixel plus bilinear at quarter-pixel), allows up to three reference frames (last frame, golden frame, alt-ref frame), and introduces the alt-ref frame concept: an invisible reference frame the encoder can synthesise to improve prediction of future frames.
Loop filter and entropy coder. VP8 has one in-loop deblocking filter and uses a boolean arithmetic coder (a simplified arithmetic coder) instead of H.264's CABAC. The arithmetic coder is a deliberate compromise: it is easier to license around the H.264 patents and slightly less efficient in pure compression terms.
In practice VP8 is roughly equivalent to H.264 Baseline Profile on most content: comparable compression on simple sources, somewhat worse on hard sources (complex motion, high detail, low bitrate), and noticeably worse than H.264 High Profile (which adds 8×8 transforms, CABAC, and B-frames that VP8 lacks). In a streaming context this gap is real but small; in a WebRTC context, where everyone codes at low bitrates and short GOPs, it does not matter and VP8 routinely matches H.264 Baseline at the same Mean Opinion Score (MOS).
VP8 has no B-frames (bidirectional predicted frames that reference both past and future frames). The alt-ref frame mechanism partially compensates by giving the encoder an invisible reference to predict from, but the missing B-frames are the main reason VP8 falls short of H.264 High Profile.
How VP9 Works: The Five Wins Over VP8
VP9 keeps the hybrid block-based framework and rebuilds every stage of it. VP9 gets its 40–45% bitrate saving over VP8 from five places (Google's design target was 50%).
1. Superblocks instead of macroblocks
VP9 throws out the fixed 16×16 macroblock and replaces it with a superblock of 64×64 pixels, recursively split into smaller prediction blocks down to 4×4 through a quadtree. Compared to HEVC, where the Coding Tree Unit (CTU) quadtree splits only into square coding units and rectangular shapes appear one level down as prediction units, VP9 allows rectangular partitions at every level (a 64×64 superblock can split into two 64×32 or 32×64 halves, for instance), which helps on content with strong horizontal or vertical motion such as panning shots and on-screen text.28
Picture the same chessboard-of-subdividable-squares from HEVC, with the extra freedom that any square can also be cut into two rectangles instead of four quadrants. On flat regions (sky, walls, gradients) VP9 uses one 64×64 block and saves bitrate; on textured or moving regions it subdivides as needed.
2. Ten intra-prediction modes, with smarter primary modes
Counter-intuitively, VP9 has only 10 intra-prediction modes, fewer than HEVC's 35. VP9's modes are designed to cover the most common edge directions efficiently (vertical, horizontal, several near-diagonal, plus DC and TM (TrueMotion) average modes) and combine with the rectangular partitioning above to match real-content edges with fewer mode bits. The mode-count gap is one of the technical reasons VP9 falls slightly short of HEVC in head-to-head Bjøntegaard Delta Bitrate tests, by a margin most measurements put in the 5–15% range depending on content and bitrate.7
3. Better motion: eighth-pixel vectors, three references, larger search
VP9 sharpens VP8's motion compensation in three places. Motion vectors are precise to 1/8 of a pixel (against VP8's 1/4-pixel and H.264's 1/4-pixel), and the interpolation filter is 8-tap for luma rather than VP8's 6-tap. Each frame keeps eight reference slots and picks three active references (last, golden, alt-ref), and a block can be predicted from one of them or from two at once (compound prediction, which VP8 lacks), including the alt-ref frame inherited from VP8 and used much more aggressively in VP9. See Inter-Frame Coding And Motion Estimation for the broader story.
4. Bigger transforms, a new asymmetric transform
VP8 used only a 4×4 transform. VP9 adds 8×8, 16×16, and 32×32 integer transforms, plus an Asymmetric Discrete Sine Transform (ADST) for blocks where the residual is expected to have asymmetric energy distribution (the edge of a moving object, for instance). The 32×32 DCT on large flat areas is a substantial win on UHD content for the same reason CTUs are: bigger transforms compress smooth gradients far better. We cover the transform-coding family in Transform Coding: DCT, ADST, Integer Transforms.
5. Tile-based parallelism
VP9 was designed for multi-core decoding from day one. The picture is divided into tile columns that can be encoded and decoded in parallel (tile rows also exist in the bitstream, but they are not independent). Tile columns plus frame- and row-level threading let a VP9 decoder use eight or more CPU cores at 4K60, and on the encoder side libvpx adds row-based multi-threading (-row-mt 1 in FFmpeg). VP8's serial, macroblock-by-macroblock decoder cannot do this.9
The cost of all five wins is encoder complexity. libvpx-vp9 at its quality-balanced -speed 1 preset is roughly 3–8× slower than libvpx-vp8 at comparable quality settings, and 5–15× slower than libx264 at comparable presets.9 Decoder complexity is more modest: roughly 2× the work of a VP8 decoder, comfortably within the budget of any 2026 device.
Figure 2. The VP9 encoder pipeline. The five stages are the same as in VP8 (predict, transform, quantize, code, filter), but each stage was redesigned for roughly twice the compression, at the cost of 3–8× the encoder CPU.
The exact compression number — and where it falls short
Independent academic measurements have repeatedly placed VP9 about 40–45% ahead of VP8 (slightly under Google's "50%" target on average), about 5–15% behind HEVC at matched objective quality on HD and UHD content, and about 25–30% behind AV1.7 On low-bitrate, low-resolution content the gap to HEVC narrows or reverses. VP9 sometimes beats HEVC at sub-720p sub-1 Mbps streaming because its rectangular intra partitions match small-block edges better than HEVC's square-only intra partitions. On high-bitrate, high-resolution content, VP9 trails HEVC by closer to 15%.
In short: VP9 is HEVC-class compression with a permissive licence and slower encoders.
VP9 Profiles And Levels — What You Actually Ship
VP9 has a much smaller profile and level matrix than HEVC because it covers a narrower set of use cases. Four profiles are defined, named Profile 0 through Profile 3, each adding a feature on top of the previous one.2
| Profile | Bit depth | Chroma subsampling | Where it ships in volume |
|---|---|---|---|
| Profile 0 | 8-bit | 4:2:0 | The default for SDR streaming up to 4K; the only profile YouTube serves below 4K HDR; the WebM-on-the-web default. |
| Profile 1 | 8-bit | 4:2:2 / 4:4:4 | Niche professional and intermediate workflows. |
| Profile 2 | 10/12-bit | 4:2:0 | The HDR profile — YouTube HDR, broadcast HDR experiments, archive HDR. |
| Profile 3 | 10/12-bit | 4:2:2 / 4:4:4 | High-bit-depth professional capture and archival. |
Table 1. The four VP9 profiles. Profile 0 is the practical default for streaming; Profile 2 is the HDR profile and the only other one with meaningful hardware decode coverage.
VP9 levels (Level 1 through Level 6.2) cap resolution, frame rate and bitrate in roughly the same shape as HEVC levels, scaled to VP9's bitstream syntax. Level 5.1 (4K60, 120 Mbps cap) is the practical default for 4K streaming. Level 6.1 caps 8K60. Most consumer hardware that advertises VP9 support is Level 5.1 capable; some 2024–2026 TVs and the latest Android flagships support Level 6.1.
The WebM container is the natural delivery wrapper for VP9. WebM is a profile of Matroska (MKV) restricted to VP8, VP9 or AV1 video and Vorbis or Opus audio. Browsers serve VP9 inside WebM by default; smart TVs and Android devices accept VP9 inside MP4 as well. See Containers: MP4, FMP4, MKV, WebM, MOV, MPEG-TS for the container side of the story.
VP8 In WebRTC, And Why It Still Matters In 2026
If you ship real-time video of any kind (video calling, conferencing, customer-support video, live shopping, telemedicine, e-learning, online tutoring), you ship VP8 today, because RFC 7742 requires every WebRTC browser to implement it as a mandatory-to-implement (MTI) video codec.4 H.264 Constrained Baseline is the other MTI codec. VP9 and AV1 show up more and more as negotiated optional codecs, but VP8 is what the browsers will always have in common when the SDP offer/answer exchange finds nothing better. We cover the negotiation itself in the SDP offer/answer deep dive, and how an SFU uses VP8 and VP9 layers in simulcast, SVC and the SFU.
Inside WebRTC, VP8 wins mostly for reasons that have little to do with compression. VP8's encoder runs in real time on any device built after 2012. Its bitrate adapts cleanly to changing network conditions, since the encoder has no B-frames to reorder and no pyramid GOP structure to recover after a packet loss. Its temporal scalability modes (later reused by VP9 and AV1) let an SFU (Selective Forwarding Unit) drop frames per participant when the network shifts, which is the conferencing advantage that made VP8 attractive in the first place. The licence cost is zero.
The trade-off: VP8 compresses only about as well as H.264 Baseline. On modern conferencing hardware that can afford H.264 High Profile or VP9, moving off VP8 saves measurable bandwidth at the same MOS. So most production WebRTC stacks offer VP8 as the floor codec, negotiate up to H.264 / VP9 / AV1 when the peer supports it, and fall back to VP8 only when SDP negotiation produces no other match. That still happens routinely on calls that cross browsers, platforms and versions. When we build a product like this as a video conferencing app development project, we set the codec fallback order explicitly and test it across browsers before launch.
Browser And Hardware Decode Support In 2026
In 2026, VP8 and VP9 support is close to universal in browsers, while hardware decode is strongest on Android.
Browsers. VP8 software decode ships in every Chrome, Firefox, Safari, and Edge in 2026 because of the WebRTC MTI requirement. Every major browser also ships VP9 software decode. Safari added VP9 in iOS 14 / macOS Big Sur (2020), Edge supports VP9 on every platform Chromium does, and Chrome and Firefox have shipped VP9 since 2013.2
Hardware decode. VP9 hardware decode is universal on Android SoCs made after 2016 (Qualcomm Snapdragon 820+, MediaTek Helio P-series and later, Samsung Exynos 8895 and later, all Google Tensor) and on smart-TV chipsets made after 2015 (most Samsung, LG, Sony, Vizio, TCL, and Hisense models). On iOS and macOS, Safari has played VP9 since Safari 14 (2020), and WebKit supports VP9 in WebRTC. Apple does not publish which chips decode VP9 in hardware, so test playback and battery drain on the oldest iPhone you support.10 On Windows, every modern Intel, AMD, and Nvidia GPU has VP9 hardware decode. The one sizeable gap is Chrome on Linux: VP9 software decode works everywhere, but hardware decode is patchy because VA-API support is inconsistent. On the server side, FFmpeg 8.0 (August 2025) added a vendor-neutral Vulkan path for VP9 decode acceleration.12
VP8 hardware decode is rarer. The use case is real-time WebRTC, where encoding is the bottleneck and bitrates are usually low enough for CPU decode. Most modern Android SoCs include VP8 hardware decode alongside VP9, but VP8 hardware encode is unusual: most devices encode VP8 on the CPU through libvpx.
Common Pitfalls (Read Before Encoding)
Here are the mistakes we keep seeing video teams make when they adopt VP8 or VP9.
Pitfall: shipping VP9 in a
video/mp4MIME type without checking the player. VP9 inside MP4 (video/mp4; codecs="vp09.00.10.08") works on Chrome, Edge, Firefox, and Android. Safari added VP9 later than the other browsers, and its support has varied by OS version and container. Test both the WebM and the MP4 path on Safari before you commit to one container for both ladders.Pitfall: encoding VP9 at libvpx defaults and accepting the result. The default
-speed 0in libvpx-vp9 is extremely slow and gives only slightly better quality than-speed 1or-speed 2. Most production ladders use-speed 1for 2-pass VOD and-deadline realtimewith-speed 6or higher for live, with-row-mt 1and-tile-columnsset to roughlylog2(width / 256). Without those flags, libvpx-vp9 can run 20× slower than it needs to.Pitfall: forgetting alt-ref frames. The biggest quality knob on libvpx-vp9 is
auto-alt-ref 1pluslag-in-frames 25, which turns on the VP9 alt-ref mechanism. Encoders that disable alt-ref (or run withlag-in-frames< 12) lose 3–10% of compression.9Pitfall: assuming VP9 Profile 2 is "VP9 HDR ready". VP9 Profile 2 covers 10/12-bit 4:2:0, the bit depth and chroma layout HDR needs. The HDR transfer function (PQ or HLG), primaries (BT.2020), and mastering display metadata still have to be signalled in the bitstream and container. See The Complete Guide To HDR: HDR10, HDR10+, Dolby Vision, HLG.
Pitfall: deploying VP9 to escape AV1's encoder cost without checking AV1's hardware decode footprint. AV1 hardware decode now ships in current Intel, AMD and Nvidia GPUs, Apple A17 Pro and M3 and later, Google Tensor phones, and most recent Android flagships and TVs. VP9 hardware decode still covers more of the installed base, but the gap shrinks with every device cycle, so measure your own audience before you decide. If you pick VP9 over AV1 in 2026, do it because the encode cost matters today and you plan to revisit the choice. AV1 decoders are no longer the problem.
VP9 In FFmpeg: The Commands You Will Actually Use
Three commands cover roughly 90% of the VP9 encoding work most video teams do. The encoder is libvpx-vp9, the open-source reference implementation maintained by Google. All three commands work on FFmpeg 6 through 8.x built with libvpx enabled.
# 1) 2-pass VOD encode, 1080p60, target 3 Mbps, quality preset
ffmpeg -i in.mov -c:v libvpx-vp9 -b:v 3M -minrate 1.5M -maxrate 4.35M \
-pass 1 -row-mt 1 -tile-columns 2 -threads 8 -speed 4 \
-auto-alt-ref 1 -lag-in-frames 25 -an -f null /dev/null && \
ffmpeg -i in.mov -c:v libvpx-vp9 -b:v 3M -minrate 1.5M -maxrate 4.35M \
-pass 2 -row-mt 1 -tile-columns 2 -threads 8 -speed 1 \
-auto-alt-ref 1 -lag-in-frames 25 -c:a libopus -b:a 128k out.webm
This is the standard libvpx two-pass recipe. A fast first pass at -speed 4 collects per-frame statistics, and a slower second pass at -speed 1 uses them to decide where the bits go. -row-mt 1 turns on row-based multi-threading inside each tile. -tile-columns 2 lets the decoder parallelise across 4 tiles at 1080p. Rate control modes are covered in Rate Control: CBR, VBR, CRF, ABR, Capped CRF.
# 2) 4K HDR Profile 2 encode (10-bit BT.2020 PQ), single pass CRF
ffmpeg -i in_master.mxf -c:v libvpx-vp9 -profile:v 2 -pix_fmt yuv420p10le \
-crf 28 -b:v 0 -row-mt 1 -tile-columns 3 -threads 16 -speed 2 \
-auto-alt-ref 1 -lag-in-frames 25 \
-color_primaries bt2020 -color_trc smpte2084 -colorspace bt2020nc \
-c:a libopus -b:a 192k out_4khdr.webm
The -crf 28 -b:v 0 pair is constant-quality mode (similar to x265's -crf): VP9 spends whatever bitrate the picture needs to reach the target perceptual quality. -pix_fmt yuv420p10le is mandatory for Profile 2. The colour signalling (-color_primaries bt2020 -color_trc smpte2084 -colorspace bt2020nc) is what makes a compatible player display the file as HDR. Without it, you get a 10-bit SDR file.
# 3) Low-latency live encode (WebM over TCP), 720p30, single pass CBR
ffmpeg -re -i input -c:v libvpx-vp9 -b:v 2M -minrate 2M -maxrate 2M \
-row-mt 1 -tile-columns 1 -threads 4 -deadline realtime -speed 6 \
-error-resilient 1 -frame-parallel 1 -lag-in-frames 0 \
-g 60 -keyint_min 60 -auto-alt-ref 0 \
-c:a libopus -b:a 96k -f webm tcp://ingest.example.com:9000
The fixed two-second keyframe interval (-g 60 at 30 fps), -error-resilient 1, and lag-in-frames 0 keep the stream low-latency and let it recover after packet loss. -deadline realtime -speed 6 selects libvpx's real-time mode. RTMP and WebRTC ingest each need their own muxer or SDK on top of the same encoder settings. See the streaming protocol comparison matrix for the protocol side.
What Changed For VP8 And VP9 In 2025–2026
Both bitstreams are frozen, so the codecs themselves will not change. The tooling, the security situation and the competing codecs have moved. Eight facts as of November 2026:
- libvpx v1.17.0 (“Yellow Billed Duck”) is the current release, tagged in August 2026. It enables high-bit-depth input validation for VP9 by default, adds Arm Neon, AVX2 and AVX-512 optimisations and C11 support, tightens API validation, drops SVC in 2-pass encoding, and fixes buffer overflows, integer overflows and decoder crashes.16
- libvpx v1.16.0 (“Xenonetta Duck”) was finalized on 27 January 2026. It added Arm SVE2 and Neon optimisations for 12-tap filters, AVX-512 SAD functions, and per-frame and per-spatial-layer PSNR, and it fixed several heap-buffer and integer overflows. It is ABI-incompatible with 1.15, so rebuild anything that links libvpx dynamically.11
- FFmpeg 8.0 shipped on 22 August 2025, and its Vulkan video additions include Vulkan-based VP9 decode acceleration.12 FFmpeg 8.1 (“Hoare”, 16 March 2026) extended the Vulkan compute codecs to ProRes and DPX. The libvpx-vp9 commands in this article run on it unchanged.17
- libvpx is a security dependency. CVE-2023-5217, a heap buffer overflow in libvpx's VP8 encoder, was exploited in the wild in September 2023 and patched in Chrome 117. If your product encodes or decodes VP8 or VP9, track libvpx releases the way you track OpenSSL.13 In 2026 we have already seen two more high-severity libvpx heap buffer overflows: CVE-2026-1861, fixed in Chrome 144.0.7559.132, and CVE-2026-2447, fixed in Firefox 147.0.4 on 16 February 2026.1819
- AV2 is out. The Alliance for Open Media released the AV2 specification in June 2026 under the same royalty-free policy, with libavm as the reference software. The 1.0.0 bitstream specification is dated 28 May 2026. In October 2026 AOMedia reported that YouTube plans to serve AV2 on the web as soon as early 2027, but no consumer device decodes it in hardware yet. So in 2026 AV2 matters for VP9's long-term outlook, while your current ladder stays as it is.1420
- AV1 keeps taking over VP9's streaming role. In December 2025 Netflix reported that AV1 powers about 30% of its streaming and said it expects it to become its top codec.6
- WebRTC now has a third common codec family. Chrome 136 (2025) shipped H.265 (HEVC) in WebRTC on desktop, Android and WebView. VP8 and H.264 are still the mandatory pair under RFC 7742, so VP8 remains the floor every browser can fall back to.21
- “Royalty-free” has a footnote. Google's grant covers VP8 and VP9, but Sisvel has run a VP9 patent pool since March 2019. The pool licenses hardware implementations and says it does not seek royalties on encoded content.15
VP8 / VP9 Vs The Field: Codec Comparison In 2026
A fair comparison has to include the five codecs your product manager will bring up in a budget meeting.
| Criterion | H.264 | VP8 | VP9 | HEVC | AV1 |
|---|---|---|---|---|---|
| Year first shipped | 2003 | 2008 / 2010 (open) | 2013 | 2013 | 2018 |
| Compression vs H.264 | baseline | ~equal (Baseline) | ~40% better | ~50% better | ~60–70% better |
| Hardware decode, share of active devices (our rough 2026 estimate) | ~100% | partial (WebRTC stacks) | ~85% | ~95% | ~80% |
| Browser support 2026 | universal | universal (WebRTC MTI) | universal | partial: needs a hardware decoder (Safari, Chrome, Edge, recent Firefox on Windows) | near-universal: Chrome, Firefox, Edge; Safari only with AV1 hardware decode (A17 Pro, M3 and later) |
| Royalty model | MPEG LA pool, free-to-end-user internet video | royalty-free (Google grant) | royalty-free (Google grant) | 2 pools + residual | royalty-free (AOMedia) |
| WebRTC mandatory | yes | yes | optional | no | optional |
| Reference encoder | x264 / OpenH264 | libvpx | libvpx | x265 | libaom / SVT-AV1 / rav1e |
| Encoder CPU vs H.264 | 1× | ~1.5× | 5–15× | 2–10× | 5–20× |
Table 2. Codec options in 2026. VP8 is the WebRTC default, VP9 is the royalty-free middle option between H.264 and AV1, and AV1 is the new royalty-free top of the line. We keep a live version of this table at Codec Comparison Matrix.
In product reviews we read this table as follows: VP8 is unavoidable inside WebRTC; VP9 is the safe royalty-free alternative for the device tail that lacks AV1 hardware decode; AV1 is the default for everything else. VP9 will fade out gradually as AV1 hardware decode reaches the remaining older devices and AV2 (specification released June 2026) starts its own hardware cycle. WebRTC still depends on VP8, though, and that will keep both libvpx codecs maintained well into the 2030s.
Figure 3. Codec options in 2026 at a glance. VP8 and VP9 sit on the royalty-free side with universal browser coverage, and their compression efficiency is mid-pack. The hardware-decode shares are our rough estimates of the active device base, not survey data.
Frequently Asked Questions
What is VP8?
The VP8 codec is a royalty-free video codec created by On2 Technologies in 2008 and open-sourced by Google in May 2010. Its compression is roughly on par with H.264 Baseline. It ships inside WebM files and is one of the two video codecs every WebRTC browser must support.
Is VP9 better than VP8?
Yes, for compression: VP9 needs about 40–45% less bitrate than VP8 at the same quality and adds 10/12-bit HDR. VP8 is still the better choice when encoder CPU is scarce or when a WebRTC peer may support nothing else.
What is a VP8 decoder?
A VP8 decoder is the software or hardware that turns a VP8 bitstream back into pictures. Every browser has one built in, FFmpeg ships both a native vp8 decoder and libvpx, and many Android chips decode VP8 in hardware.
Is VP9 better than H.265?
Not on compression: HEVC needs about 5–15% less bitrate at matched quality. VP9 wins on licensing and browser playback; HEVC wins on hardware encode and Apple ecosystem support.
Does WebM use VP8 or VP9?
Both. WebM is a Matroska profile that carries VP8, VP9 or AV1 video with Vorbis or Opus audio, and most new WebM files use VP9.
How do I encode or transcode video to VP9?
Use FFmpeg with libvpx-vp9, for example ffmpeg -i in.mp4 -c:v libvpx-vp9 -crf 31 -b:v 0 -row-mt 1 -c:a libopus out.webm. For production ladders, use the two-pass recipe from the FFmpeg section above.
What is VP9?
The VP9 codec is Google's royalty-free video codec from 2013 and the successor to VP8. At the same quality it needs about 40–45% less bitrate than VP8. It supports 10- and 12-bit HDR through Profile 2 and is YouTube's main codec for devices that do not decode AV1.
When was VP9 released?
Google froze the VP9 bitstream in June 2013, the same month YouTube started serving VP9 video. Development began in late 2011.
What does vp09 mean in a codec string?
vp09 is the four-character code for VP9 in a MIME codecs parameter such as video/mp4; codecs="vp09.00.10.08". The fields that follow are profile, level and bit depth, so vp09.00.10.08 means VP9 Profile 0, Level 1.0, 8-bit. VP8 uses vp08.22
What is VP9 Profile 2?
VP9 Profile 2 is the 10- and 12-bit 4:2:0 profile used for HDR, YouTube HDR included. The HDR transfer function, BT.2020 primaries and mastering metadata still need to be signalled separately.
Can VP9 be used for live streaming?
Yes. libvpx-vp9 has a real-time mode (-deadline realtime with -speed 6 or higher). With -row-mt 1, tile columns and a fixed keyframe interval it encodes 720p and 1080p live, and the third FFmpeg recipe above is a working starting point.
Is VP8 still used in 2026?
Yes. VP8 is still mandatory in WebRTC under RFC 7742, so every major browser still encodes and decodes it, and libvpx v1.17.0 (August 2026) continues to maintain it.
Free download: VP8 / VP9 Cheat SheetOne-page A4 reference: the VP9 profile and level matrix, the libvpx FFmpeg recipes from this article, and the 2026 browser and hardware decode snapshot.Where Fora Soft Fits In
We have been building video products since 2005 and have shipped 250+ projects in video conferencing, video streaming, OTT and Internet TV, video surveillance, e-learning, telemedicine, and AR/VR. VP8 is part of every real-time pipeline we have delivered. Each WebRTC SFU integration ships VP8 by default as the floor codec. Our telemedicine platforms fall back to VP8 when the patient's browser is too old to negotiate H.264, and our conferencing products layer VP8's temporal scalability over an SFU to keep bandwidth in check on weak networks. VP9 is in the codec ladder of every streaming product we have built where the client wanted a royalty-free 4K rendition for the YouTube-class device tail without paying the AV1 encoder cost on day one. We have the "VP9 or skip to AV1" conversation with streaming clients several times a quarter, and the answer is rarely "AV1 only": most teams ladder all three (H.264 + VP9 + AV1) for the next two to three years. If your team is working through that decision, we are glad to talk it over. Our WebRTC developers tune VP8 and VP9 simulcast and SVC on real SFU deployments, and our video streaming development team builds the multi-codec ladders for VOD and live.
The book
Video Encoding: A Complete Guide to Video Compression
This article walks the VP9 pipeline once; the book takes each stage apart. Superblock partitioning, the ten intra modes, eighth-pixel motion and the ADST all get their own chapters, the alt-ref mechanism is worked through with the libvpx flags that switch it on, and the rate-control chapter explains why -speed 1 and -row-mt 1 change the encode by an order of magnitude. It is volume 1 of the seven-volume Fora Soft Video Engineering Handbook, 1,522 pages, written by Nikolay Sapunov, CEO at Fora Soft.
VP9, AV1, or both — which belongs in your ladder?
Send us the device and browser mix you have to support and your current encoder settings, and we will tell you what a VP9 rung would cost you to produce and how much of your audience it actually reaches that AV1 does not. For real-time products we do the same on the negotiation side: which share of your calls still falls back to VP8, and what H.264 High Profile or VP9 would save at the same MOS. You get the numbers on your own traffic. 250+ video products shipped since 2005.
See our case studies
Case study
TransLinguist
Remote simultaneous interpretation with an interpreter marketplace, built on WebRTC. Every call negotiates video in the browser, so VP8 is the floor the SDP exchange lands on when nothing better is common — exactly the RFC 7742 situation this article describes.
Case study
Worldcast Live
Concert streaming at 1.5 Gb/s with 0.4–0.5 s latency to 10,000 simultaneous viewers on a WebRTC and Kurento stack. At that bitrate the codec choice is a line in the bandwidth bill, and the real-time constraint is why the encoder settings look nothing like the two-pass VOD recipe above.
Case study
Speed.Space
Live 1080p at 8 Mb/s for production clients including Netflix, EA and HBO. This is where the “VP9 rung or straight to AV1” question gets decided in practice: which renditions the ladder carries, and which device tail each one is there to reach.
What to read next
Codecs
H.264 / AVC: The Workhorse Of The Internet
The codec VP8 was built as a royalty-free counterpart to, and the other mandatory codec in every WebRTC browser.
Codecs
Comparison Table: MPEG-2, H.264, H.265, VP9, AV1, VVC
The live version of the table above, with every production codec on every dimension a product team weighs.
Tooling
Containers: MP4, fMP4, MKV, WebM, MOV, MPEG-TS
WebM is the wrapper VP8 and VP9 ship in — and the MP4 path is the one to test on Safari before you commit.
Tooling
Encoder Implementations: x264, x265, libvpx, SVT-AV1, rav1e
libvpx is the reference encoder behind both codecs here; this is how it compares with the rest on speed and quality.
-
Wikipedia. "VP8." Accessed 2026-05-16. https://en.wikipedia.org/wiki/VP8 ↩↩↩
-
Wikipedia. "VP9." Accessed 2026-05-16. https://en.wikipedia.org/wiki/VP9 ↩↩↩↩↩
-
Google Official Blog / Wikipedia. "Google acquisition of On2 Technologies." Closed 19 February 2010 for USD 124.6 million. https://en.wikipedia.org/wiki/On2_Technologies ↩
-
IETF. "RFC 7742: WebRTC Video Processing and Codec Requirements." March 2016. https://datatracker.ietf.org/doc/html/rfc7742 ↩↩
-
Bitmovin. "Video Developer Report 2025." Streaming codec adoption survey, conducted September–December 2024. https://bitmovin.com/video-developer-report-2025/ ↩
-
Netflix Tech Blog, via Streaming Media. "Netflix says AV1 codec now powers 30% of Netflix VOD streaming." December 2025. https://www.streamingmediablog.com/2025/12/netflix-av1.html ↩↩
-
Grois, D., et al. "Comparison of Compression Efficiency between HEVC/H.265, VP9 and AV1 based on Subjective Quality Assessments." IEEE PCS 2018. https://ieeexplore.ieee.org/document/8463294 ↩↩↩↩
-
Paul, S., et al. "Speeding up VP9 Intra Encoder with Hierarchical Deep Learning Based Partition Prediction." IEEE Transactions on Image Processing, 2020. https://ieeexplore.ieee.org/document/9151395 ↩
-
WebM Project. "FFmpeg VP9 Encoding Guide." https://wiki.webmproject.org/ffmpeg/vp9-encoding-guide ↩↩↩↩
-
Bitmovin / Wowza technical references on VP9 hardware decode support. https://bitmovin.com/blog/vp9-codec-status-quo/ ↩
-
WebM Project discussion group. "libvpx v1.16.0 Xenonetta Duck Finalized." 27 January 2026. https://groups.google.com/a/webmproject.org/g/webm-discuss/c/NsLvY2mRj7M ↩
-
Phoronix. "FFmpeg 8.0 Released With OpenAI Whisper Filter, Many Vulkan Video Improvements." 22 August 2025. https://www.phoronix.com/news/FFmpeg-8.0-Released ↩↩
-
NIST National Vulnerability Database. "CVE-2023-5217: heap buffer overflow in vp8 encoding in libvpx." September 2023. https://nvd.nist.gov/vuln/detail/CVE-2023-5217 ↩
-
Alliance for Open Media. "Alliance for Open Media Releases AV2 Codec." June 2026. https://aomedia.org/press%20releases/Alliance-for-Open-Media-Releases-AV2-Codec/ ↩
-
Sisvel. "Sisvel VP9/AV1 pools re-launch and editorial response." March 2020. https://www.sisvel.com/insights/sisvel-vp9-av1-pools-re-launch-and-editorial-response/ ↩
-
libvpx Git repository, tag v1.17.0 “Yellow Billed Duck” (August 2026); release notes in the v1.17.0-rc1 announcement, WebM Project discussion group, 24 June 2026. https://chromium.googlesource.com/webm/libvpx/+refs and https://groups.google.com/a/webmproject.org/g/webm-discuss/c/5tAFIvEQveo ↩
-
Ubuntu Handbook. "FFmpeg 8.1 Released with Rockchip & JPEG-XS Support." 17 March 2026 (release 16 March 2026, codename Hoare). https://ubuntuhandbook.org/index.php/2026/03/ffmpeg-8-1-rockchip-jpeg-xs/ ↩
-
Debian Security Tracker. "CVE-2026-1861: heap buffer overflow in libvpx in Google Chrome prior to 144.0.7559.132." https://security-tracker.debian.org/tracker/CVE-2026-1861 ↩
-
Mozilla Foundation Security Advisory 2026-10. "CVE-2026-2447: Heap buffer overflow in libvpx." 16 February 2026. https://www.mozilla.org/security/advisories/mfsa2026-10/ ↩
-
Alliance for Open Media. "Alliance for Open Media Sees Rapid AV2 Adoption Across the Video Ecosystem Just Four Months After Release." 22 October 2026. https://aomedia.org/press%20releases/AOMedia-Rapid-AV2-Adoption-Across-the-Video-Ecosystem/ ↩
-
Chromium blink-dev. "Intent to Ship: H265 (HEVC) codec support in WebRTC." Chrome 136, 3 March 2025. https://groups.google.com/a/chromium.org/g/blink-dev/c/3h8lL8a377c ↩
-
MDN Web Docs. "Codecs in common media types: VP9." https://developer.mozilla.org/en-US/docs/Web/Media/Guides/Formats/Codecs_parameter ↩
