Snowflake's AI research team released the Arctic Embed model family.

Arctic Embed is a family of compact, efficient text embedding models tuned specifically for high-quality document retrieval performance relative to their size.

Hosted on Hugging Face under the Snowflake organization and on GitHub at Snowflake-Labs/arctic-embed.

The first Arctic Embed models were released in April 2024, with the multilingual Arctic Embed 2.0 versions following in December 2024.

It targets enterprises needing strong retrieval quality with lower serving cost than larger embedding models.

Compatible with the Hugging Face sentence-transformers library.

Released under the Apache 2.0 license by Snowflake; fully open weights with no gating.