Skip to main content
Google says AI rewrote a video decoder and made it 20–30% faster in tests

Google says AI rewrote a video decoder and made it 20–30% faster in tests

CryptoSlate
16 lis 2026, 21:11Cryptoslate
Oryginał
1 min read

Google reports its Antigravity AI system successfully rewrote the libgav1 video decoder in Rust, achieving 20-30% faster performance than the original C++ and hand-optimized assembly code on sample video clips.

The disclosure, made in a November 13 engineering update, followed an initial port that ran approximately four times slower than the C++ version. Google reports achieving around 170,000 lines of Rust code and 100% bit-exact parity across a broad conformance test suite, meaning the rewritten decoder matched the original's output in those tests.

Performance reversal after profiling

The performance improvement came after /teamwork received profiling tools and a specific goal of maximizing speed on sample clips. Google reports the speed advantage on both x86_64 and ARM architectures, though the disclosure omits key benchmark settings and hardware models used in testing.

Separating behavior from speed

The migration illustrates two distinct engineering challenges: preserving an established program's behavior and making the replacement fast enough for practical use. Google describes a forthcoming general rewrite pattern that schedules each module only after its dependencies compile, then tests leaf modules against the original test suite while keeping those checks intact.

Keeping those checks intact gives the agents a fixed target while they change the implementation. Profiling then provides a separate feedback loop for finding where the new version spends its time.

Google engineering team

No production deployment yet

The libgav1 project repository still documents a C++17 and CMake build, including architecture-specific optimization switches. The November report does not specify processor models, compiler settings, clip identities, thread counts or timing distributions. It also does not identify a reproduction package for the Rust benchmark or announce that the rewritten decoder has entered production.

This leaves a concrete next test for the disclosure: a specified Rust revision, comparison build and video corpus would allow other developers to examine whether the reported advantage holds under their own conditions.

Z.ai
Article written by AI
Z.ai GLM-4.5-FlashFree
Free AI articles150 of 150
No limit after signing in with Telegram