Embedding Models
mixedbread embed is our flagship embedding model family. Enjoy easy access and stellar performance that can help you elevate your retrieval pipeline. Use embeddings for search, classification, recommendation, and other impactful tasks.
What's new in the mixedbread embed model family?
With the recent release of our tasty English embedding model mxbai-embed-large-v1, the mixedbread embed family has finally seen the light of day!
The model family now includes:
| Model | Status | Context Length | Dimension | MTEB Average |
|---|---|---|---|---|
| mxbai-embed-large-v1 | API available | 512 | 1024 | 64.68 |
| mxbai-embed-2d-large-v1 | API Available - Research preview | 512 | 1024 (base) | 63.25 (base) |
Coming soon: We are currently working on specialized models to extend the family! Please feel free to contact us for more information.
Why mixedbread embeddings?
mixedbread embed is a powerful, size-efficient embedding model family - and the best part is, it's fully open-source! The new mxbai-embed-large-v1 model outperforms other similarly sized open models currently available on the MTEB benchmark, and its performance even surpasses that of current closed-source models:
| Model | Avg (56 datasets) |
|---|---|
| mxbai-embed-large-v1 | 64.68 |
| bge-large-en-v1.5 | 64.23 |
| jina-embeddings-v2-base-en | 60.38 |
| Proprietary Models | |
| OpenAI text-embedding-3-large | 64.58 |
| Cohere embed-english-v3.0 | 64.47 |
What's the benefit of using the API?
You can start using our open-source model, but we offer certain advantages and features for the models provided through our API, with more to come in the future. For example, the API-exclusive version offers better performance for int8-quantization. It can more accurately map float32 to int8 values because we generated calibration data for this feature using over 50 million data samples.
How can you get started using mixedbread embeddings yourself?
Our model is extremely easy to use with your existing search stack. Learn more in the following sections about how you can use our models.