Microsoft's Bing team open-sources "Harrier" embedding model

Microsoft’s Bing team open-sourced Harrier, an embedding model.

Harrier supports more than 100 languages and a 32,000-token context window.

It was trained on over two billion examples plus GPT-5 synthetic data.

The team says Harrier ranked top on the multilingual MTEB v2 benchmark and outperformed OpenAI and Amazon models.

Three sizes—harrier-oss-v1-27b (27B), 0.6B, and 270M—are available on Hugging Face under the MIT license; Microsoft plans to integrate the tech into Bing and agent grounding services.