LLM ReferenceLLM Reference

Harrier OSS v1 27B

harrier-oss-v1-27b

Open Source

About

Microsoft's largest open-source multilingual text embedding model, released March 30, 2026. Decoder-only architecture based on Gemma 3, with last-token pooling and L2 normalization. Produces 5,376-dimensional embeddings. Ranked #1 on multilingual MTEB-v2 benchmark (score: 74.3). Supports 94+ languages. MIT license. Available on HuggingFace and Azure AI Foundry.

Harrier OSS v1 27B has a 32K-token context window.

Capabilities

VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode ExecutionPrompt CachingBatch APIAudioFine-tuning

Rankings

Specifications

FamilyHarrier
Released2026-03-30
Parameters27B
Context33K
Specializationembedding
LicenseMIT

Created by

Advancing the state-of-the-art in AI and computing.

Redmond, Washington, United States
Founded 1991
Website