Nomic Embed Text v2

Embedding modelOpen weightsCard summarized

Multilingual Mixture of Experts Text Embeddings

Source: Training Sparse Mixture Of Experts Text Embedding Models (technical report, arXiv 2502.07972 v3) and Hugging Face model card

Use it via API

Self-host only: no published API price.

Code example Compare

Where to call it

Self-host (weights)nomic-ai/nomic-embed-text-v2-moe
GitHub (training code)nomic-ai/contrastors

Supports

No feature list in the maker's docs.

Full feature support

Good for

Multilingual retrieval (search queries vs documents)
RAG retrieval where latency and memory matter: 305M active parameters
Storage-constrained indexes using 256-dimension vectors

Watch out for

Short context
Uneven quality across languages
Higher resource needs than a dense model of the same active size

Overview

Nomic · 01Text

Embed Text v2

Multilingual Mixture of Experts Text Embeddings

Typical price

—

No published price

Dimensions: 768
Max input: 512 tokens
Modes: text embeddings, multilingual retrieval
Released: 11 Feb 2025

Nomic Embed v2 is a multilingual text embedding model that Nomic describes as the first general-purpose Mixture of Experts (MoE) text embedding model. It has 475M total and 305M active parameters, supports about 100 languages, outputs 768-dimension vectors that can be truncated to 256, and was trained on 1.6B filtered pairs. Weights, training code and training data are released. report abstract, §4; model card

nomic-ai/nomic-embed-text-v2-moe

Typical price: —
Dimensions: 768 (Matryoshka: can be truncated to 256)
Max input: 512 tokens
Languages: ~100 languages
Weights: Open · Apache 2.0
Released: 11 Feb 2025

text embeddings
multilingual retrieval