Our Take
Meta solved a real engineering problem (AV1 power draw and device coverage) but shipped no independent benchmarks proving the 20% bitrate claim holds in live calls or that users see measurable quality gains.
Why it matters
Billions of people on slow networks rely on these apps; lower bitrate codecs mean fewer dropped calls and better video in emerging markets. This matters now because RTC deployment at Meta's scale forces choices other teams will face when scaling newer codecs.
Do this week
Video team: audit your codec's power profile on the lowest-tier Android devices you support before committing to encoder adoption; Meta's 14% power overhead forced a custom low-complexity build.
Meta deployed AV1 to majority of mobile devices in Messenger and WhatsApp
Meta completed a multi-year rollout of AV1 video codec for real-time calls, moving from high-end devices only (introduced in 2023) to most Android and iOS phones by late 2025. The motivation is straightforward: AV1 delivers the same visual quality at lower bitrates. In offline tests on low-end and mid-range devices, Meta observed at least 20% bitrate reduction compared to H.264/AVC under their product settings (company-reported). For calls at 100 kbps—a typical constraint in emerging markets—AV1 produced noticeably clearer video than H.264 in side-by-side tests.
AV1 includes encoding tools tailored to screen content (palette mode and intra-block copy) that improve compression of text and repetitive patterns, addressing a long-standing weakness of older codecs. Real-time communication systems, unlike video-on-demand services, must maintain end-to-end latency below 300 milliseconds and handle sudden network fluctuations without freezing the video. Any new codec must solve these constraints simultaneously.
Building a low-complexity encoder
The largest obstacle was power consumption. Open-source AV1 encoders consumed 14% more power than H.264/AVC baseline on a Pixel 8 device during a call. Meta adopted an internal low-complexity encoder with comparable power to H.264, then developed an ultra-low-complexity preset for entry-level phones. The encoder automatically adjusts its preset based on device capabilities, enabling AV1 deployment across a much broader range of hardware.
For decoding, Meta A/B tested open-source options and selected dav1d for superior power efficiency and real-time performance on low-end devices. Initial trials on weak devices showed video freezes and sync issues until the decoder switch reduced computational load.
Solving binary bloat
Adding AV1 encoder and decoder libraries increased the app binary by 1.7 MB (600 kB compressed), a severe constraint for a platform serving billions of users. A 600 kB increase could consume an entire year's binary budget for a large organization and delay app startup and incoming call setup. Meta pursued direct optimizations: removing unused tools from the codec library (e.g., removing quantization matrix tools freed 60 kB), contributing size reductions to the dav1d decoder project, and sharing codec libraries across features like video message transcoding. Platform-level hardware codec support also reduced the need to bundle additional libraries.
Machine learning approach to device coverage
Android fragmentation made manual device selection unreliable. Simple heuristics based on RAM, release year, or OS version failed to predict AV1 performance consistently. Meta built an ML-based device eligibility framework that uses large-scale real-world performance metrics collected through logging to assign an rtc_score to each device. This score determines whether AV1 can run efficiently on that hardware.
The model launched in August 2025 (V1.1) and expanded AV1 traffic across more devices, generating a larger dataset that fed into Model V2, which introduced a two-tier approach differentiating higher-end and lower-end phones. Meta expects iterative refinement to continue as AV1 traffic grows.
Real-time compression gains matter most where bandwidth is scarce
In real-world networks, especially in emerging markets, RTC bitrates typically range from 10 kbps to 400 kbps. Maintaining good video quality below 100 kbps has been consistently difficult. A 20% bitrate reduction means people on slower or limited networks can either enjoy better video quality at the same bitrate or use less data and battery for the same quality. The deployment also signals that newer codecs can ship on mainstream devices without requiring custom hardware, removing one excuse for codec stagnation in real-time systems.
However, the 20% bitrate savings remain company-reported and unvalidated by independent benchmarking in live-call conditions. No published A/B test results yet show whether end users perceive the quality improvement or whether network efficiency gains translate to fewer dropped calls or longer talk time on battery.
Audit codec power before scaling; expect binary and latency tradeoffs
Teams considering adoption of newer video codecs should profile power consumption and memory overhead on the lowest-tier devices they support before committing to encoder integration. Meta's 14% power penalty forced custom codec work; teams without in-house codec expertise should calculate the cost of that effort against the bitrate gain. Binary size budgets are real constraints in consumer mobile applications, and codec libraries are not exempt. Device eligibility systems based on lab specs will fail at scale; consider telemetry-driven ML approaches to avoid deployment surprises. Finally, real-time systems cannot tolerate the latency overhead of quality-focused multi-pass encoding; optimize for streaming constraints first, then quality.