Nomic Embed

Nomic Embed text v1 is an 8192-token-context text embedding model from Nomic AI, released in early 2024. It was designed as the first fully reproducible long-context embedder: the model weights, the training code, and the entire curated training dataset were all published under the Apache 2.0 license, a level of openness few competitors match. On the MTEB benchmark it scored 62.39 and on the long-context LoCo benchmark 85.53, beating OpenAI's text-embedding-ada-002 and text-embedding-3-small on short and long tasks.

The honest picture: it is an English-focused encoder from the BERT family era, so multilingual retrieval and very recent embedding architectures have moved past it. It also requires task-instruction prefixes (search_query: / search_document:) to perform well, which adds integration friction. But for teams that want a transparent, self-hostable embedder with no API dependency, it remains one of the most auditable options available.