Overview
Nomic-embed-text is the first fully reproducible, open-source text embedding model with 8,192 context length that outperforms both OpenAI Ada-002 and text-embedding-3-small on short and long context benchmarks.
Key Features
- Fully Open Source: Training code, model weights, and complete training data released
- Apache 2.0 License: Free for commercial use
- 8,192 Context Length: Long context support
- Reproducible: Complete replication possible with released data and code
- High Performance: Outperforms OpenAI models on MTEB benchmarks
Model Versions
V1 (nomic-embed-text-v1)
- First fully reproducible embedding model
- 8,192 context length
- Trained on weakly related text pairs and high-quality labeled datasets
- English-focused
V1.5 (nomic-embed-text-v1.5)
- Matryoshka Representation Learning support
- Flexible embedding dimensions
- Trade-off between size and performance
- Minimal performance reduction with smaller dimensions
V2 (nomic-embed-text-v2)
- Mixture-of-Experts (MoE) Architecture: First MoE text embedding model
- Multilingual: Trained on 1.6 billion contrastive pairs across ~100 languages
- Expanded Dataset: Broader multilingual coverage
- Production-Ready: Optimized for real-world applications
Training Approach
- Stage 1 - Unsupervised Contrastive: Training on weakly related text pairs from StackExchange, Quora, Amazon reviews, news articles
- Stage 2 - Fine-tuning: Leverages high-quality labeled datasets including search queries and web search answers
Access Methods
- Hugging Face: Direct model download and inference
- Ollama:
ollama pull nomic-embed-text
- Nomic API: Managed API endpoint
- LlamaIndex Integration: Native support
- Qdrant Integration: Built-in connector
Use Cases
- Long-context semantic search
- Multilingual retrieval applications
- Document embedding and clustering
- RAG systems requiring long context
- Research requiring reproducibility
- Outperforms OpenAI text-embedding-ada-002
- Competitive with text-embedding-3-small
- Strong performance on both short and long context tasks
- Excellent multilingual capabilities (v2)
Pricing
Free and open-source under Apache 2.0 license. No licensing costs. Nomic API offers managed hosting with usage-based pricing for convenience.
