Model4d ago
Harrier Embedding Model Tops Multilingual Benchmark
Harrier tops the multilingual MTEB v2 benchmark, scoring 1,200 points across 100+ languages. Microsoft’s Bing team releases the model under an open‑source license, enabling developers to embed text in any of the supported tongues. The architecture blends transformer layers with a language‑agnostic tokenization scheme. Engineers can now deploy high‑quality multilingual embeddings without training from scratch.