A new WASM SIMD implementation of TurboQuant vector quantization brings 3-bit-per-dimension compression to browsers and Node.js. Based on Google Research's ICLR 2026 paper, the teamchong/turboquant-wasm library achieves near-optimal distortion rates while enabling fast dot product calculations without full decompression. The library targets developers building vector search, image similarity, and 3D graphics applications.

Key technical features include relaxed SIMD instructions mapped to f32x4.relaxed_madd operations, SIMD-vectorized QJL sign packing, and a TypeScript API with encode/decode/dot operations. The implementation maintains bit-identical output with the original Zig version and includes golden-value tests verifying inner product preservation. Compression achieves approximately 4.5 bits per dimension payload with 6x compression ratios.

Browser compatibility requires Chrome 114+, Firefox 128+, Safari 18+, or Node.js 20+ for relaxed SIMD support. The library demonstrates practical applications through live demos including vector search, image similarity matching, and 3D Gaussian Splatting compression running directly in browsers. Developers can install via npm and initialize with configurable dimensions and seeds for reproducible results.