News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgNomic Embed

Nomic Embed

Open-source long-context text embedding model

Nomic Embed is an open-source text embedding model developed by Nomic AI. Released in 2024, it supports context lengths up to 8192 tokens and uses Matryoshka Representation Learning to allow flexible embedding dimensions. The model and its full training data are publicly available under an Apache 2.0 licence.

Share on WhatsApp

  • Supports context lengths up to 8192 tokens
  • Released under Apache 2.0 licence with open data
  • Uses Matryoshka Representation Learning for flexible dimensions
  • Outperforms OpenAI Ada-002 on MTEB benchmark
  • v1.5 adds multimodal text-vision alignment

What it is

Nomic Embed is a family of text embedding models created by Nomic AI. The initial version, nomic-embed-text-v1, was released in February 2024. It is designed to convert text into vector representations for tasks such as search, clustering, and classification. The model supports a context window of 8192 tokens, allowing it to process longer documents than many earlier embedding systems. A later version, v1.5, added multimodal capabilities by aligning text embeddings with a vision model. The project emphasises reproducibility by releasing the full training code, weights, and datasets.

How it came about

The model was introduced in a technical report submitted to arXiv in February 2024 by Zach Nussbaum, John X. Morris, Brandon Duderstadt, and Andriy Mulyar. The team aimed to create the first fully reproducible, open-source embedding model with long context support. They released the training code and data to allow others to replicate the results. The subsequent v1.5 update extended the model to support multimodal inputs, aligning text and vision embeddings within the same space.

Inside the model

The architecture is based on a BERT encoder model. Training involves a multi-stage pipeline. The first stage uses unsupervised contrastive learning on weakly related text pairs, such as question-answer threads from forums and title-body pairs from reviews. The second stage fine-tunes the model on high-quality labelled datasets, including search queries and answers. The model employs Matryoshka Representation Learning, which allows developers to truncate embedding dimensions with minimal performance loss. It also uses dynamic rotary position embeddings to handle sequence lengths beyond 2048 tokens.

What it can do

Nomic Embed performs well on the Massively Text Embedding Benchmark (MTEB) and the Long Context (LoCo) benchmark. The v1 model achieved a score of 62.39 on MTEB with 768-dimensional embeddings. The v1.5 model maintains similar performance while offering flexible dimension sizes ranging from 64 to 768. It requires task instruction prefixes, such as 'search_query' or 'clustering', to optimise embeddings for specific use cases. The model supports scaling sequence length up to 8192 tokens.

How to use it

The model weights and training code are available under an Apache 2.0 licence. Users can download the models from Hugging Face or access them via the Nomic Embedding API. The full training dataset is also public, hosted in the contrastors repository. Integration is supported through libraries like Sentence Transformers and Transformers.js. The company states that trust_remote_code flags are no longer needed for recent library versions.

Reception and debate

The model was accepted to the Transactions on Machine Learning Research journal. It is noted for outperforming proprietary models like OpenAI Ada-002 on short-context benchmarks. The release of full training data and code has been highlighted as a step towards reproducibility in the field of natural language processing.

Why it matters

Nomic Embed has contributed to the availability of high-performance, open-source embedding models. Its use of Matryoshka Representation Learning has influenced how developers manage embedding storage and retrieval efficiency. By providing complete training resources, it has set a precedent for transparency in model development.

Questions people ask about Nomic Embed

What is Nomic Embed used for?

It converts text into vector embeddings for search, clustering, classification, and retrieval-augmented generation applications. It supports both short and long context documents.

Is the training data available?

Yes, Nomic AI released the full curated training data and code. This allows users to replicate the training process entirely, which is rare for open-source models.

How does Matryoshka Representation Learning help?

It allows developers to reduce the embedding dimension size, such as from 768 to 256, with only a small drop in performance. This saves storage space and speeds up search operations.

Do I need to add prefixes to my text?

Yes, the model requires task instruction prefixes like 'search_query:' or 'classification:' before the input text. This tells the model how to structure the embedding for the specific task.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Nomic Embed yet, so the facts on this page come from the sources listed above: huggingface.co/nomic-ai/nomic-embed-text-v1.5, arxiv.org/abs/2402.01613. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.