Show HN: TurboQuant-WASM — Google's Vector Quantization Runs in the Browser at 6x Compression — 154 Points

Hacker News·medium signal

A Show HN project brings Google Research's TurboQuant (ICLR 2026) vector quantization algorithm to the browser via WASM with relaxed SIMD support. It compresses vector data 6x (7.3MB → 1.2MB) at 3 bits per dimension and enables searching directly on compressed data without decompression — encode(), decode(), and dot() with no training step or codebook required. Browser compatibility: Chrome 114+, Firefox 128+, Safari 18+. For builders, this means client-side vector search, image similarity, and 3D Gaussian Splatting compression without server roundtrips.