Nomic Embed
Apache 2.0 embedding models from Nomic AI with open training data: text v1.5 and multilingual v2 MoE, code, vision and multimodal
Overview
Nomic Embed is a family of open embedding models from Nomic AI, notable for releasing not just the weights but the training data and code (the contrastors repo). nomic-embed-text-v1.5 (about 137M parameters, 768 dimensions, Matryoshka down to 64, 8,192-token context, English) is one of the most used local embedding models, with over 97 million pulls on Ollama. nomic-embed-text-v2-moe (475M total, 305M active parameters, about 100 languages, 512-token sequences) is the multilingual successor. The family also includes nomic-embed-code (7B, for code retrieval), nomic-embed-vision-v1.5 (shares v1.5's text vector space) and ColNomic Embed Multimodal for visual document retrieval, all released between 2024 and mid 2025. No new embedding model has appeared since June 2025: Nomic now leads with the Nomic Platform, an agent product for architecture, engineering and construction firms, and its hosted Embeddings API sits inside the Atlas docs without a public per-token price. Also from Nomic: GPT4All, its local chat app, whose LocalDocs feature indexes files with Nomic's on-device embedding models.
The Verdict
Who Should Use Nomic Embed?
Best For
- Local and offline RAG on a laptop or small GPU (Ollama, GGUF)
- Teams that need fully auditable training data
- Commercial self-hosting under a permissive Apache 2.0 license
- Code search with an open 7B model
Not Ideal For
- Long multilingual documents: v2-moe stops at 512 tokens
- Buyers who want a supported, priced embedding API with an SLA
- Teams that want a vendor still shipping new embedding models
What's Great
- Apache 2.0 weights, plus open training data and code
- Matryoshka dimensions cut vector storage
- 8,192-token context on text v1.5
- Huge local adoption through Ollama
- nomic-embed-code reports beating Voyage Code 3 and OpenAI 3-large overall on CodeSearchNet
Watch Out For
- v1.5 is English only; v2-moe is multilingual but has a short 512-token window
- Ollama's default context for nomic-embed-text is 2K, not the full 8,192
- No new embedding release since mid 2025; the training repo was last pushed March 2025
- Company focus moved to its AEC agent platform
- Benchmark numbers are mostly Nomic's own
Pricing
View all features & details
Models
- nomic-embed-text-v1.5: ~137M, 768 dims (Matryoshka 512 to 64), 8,192 tokens
- nomic-embed-text-v2-moe: 475M total, 305M active, 768 to 256 dims, ~100 languages
- nomic-embed-code: 7B, Qwen2.5-Coder base; Python, Java, Ruby, PHP, JavaScript, Go
- nomic-embed-vision-v1.5: 92.9M, same space as text v1.5
- ColNomic Embed Multimodal 3B and 7B (April 2025)
- CodeRankEmbed (0.1B) and CodeRankLLM (June 2025)
Use Cases
- Local RAG and semantic search
- Image and text search in one space
- Code retrieval for coding assistants
- Research on reproducible embedding training
Real-World Usage
Adoption
- Over 97 million Ollama pulls for nomic-embed-text
- 73 models on the nomic-ai Hugging Face org
- A July 2026 roundup of community advice names nomic-embed-text the common pick for English local RAG and bge-m3 for multilingual
Company Direction
- Nomic Platform launched November 2025 for AEC firms
- Case studies with Arcadis and Aurecon; strategic funding announced September 2026
- GPT4All desktop app repo is still public and not archived
How It Compares
| Feature | Nomic Embed v1.5 | BAAI bge-m3 | Jina v5 text | OpenAI 3-small |
|---|---|---|---|---|
| License | Apache 2.0 | MIT | Check card (most Jina weights CC BY-NC) | Closed API |
| Size | ~137M | Not stated on card | 239M to 677M | Not disclosed |
| Context | 8,192 tokens | 8,192 tokens | 8K to 32K tokens | 8,192 tokens |
| Languages | English (v2-moe: ~100) | 100+ | Multilingual | Not stated by OpenAI |
| Open training data | Yes | No | No | No |
| Cost | Free to self-host | Free to self-host | API top-ups or license | $0.02 per 1M |
Sources: Nomic card · bge-m3 card · Jina embeddings · OpenAI pricing
Is this your tool? Add a badge