- Updated: March 30, 2026
- 1 min read
Microsoft AI Launches Harrier‑OSS‑v1: Breakthrough Multilingual Embedding Model Sets New SOTA
Microsoft AI has announced the release of Harrier‑OSS‑v1, an open‑source, decoder‑only multilingual text‑embedding family that pushes the state‑of‑the‑art on the original MarkTechPost article. Designed with a 32k token context window and instruction‑tuned capabilities, Harrier‑OSS‑v1 delivers top‑ranking results on the Multilingual MTEB v2 benchmark across more than 100 languages.
The model family includes several sizes, all built on a decoder‑only architecture that simplifies deployment while maintaining high performance. Its open‑source nature encourages community contributions and rapid integration into multilingual AI applications such as cross‑language search, translation assistance, and semantic retrieval.
Key highlights:
- Decoder‑only design with 32k token context window.
- Instruction‑based implementation for flexible downstream tasks.
- State‑of‑the‑art scores on Multilingual MTEB v2, surpassing previous open‑source alternatives.
- Fully open‑source license, fostering broader adoption.
For more insights on multilingual AI advancements, explore our related pages: Ubos Tech News, AI Products, and the dedicated blog post on Multilingual Embeddings.
Read the full story and download the model from the official Microsoft AI repository.