mxbai-edge-colbert-v0
AI model · Embedding model · Text → Embeddings by Mixedbread
Extremely efficient ColBERT models for edge and cloud retrieval.
mxbai-edge-colbert-v0 is a family of efficient 17M and 32M parameter ColBERT models designed for high-performance retrieval on edge devices and in the cloud.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Launched
- 2025-10-27
- Languages
- English
Category: Natural Language Processing
- Parameters
- 17M and 32M versions
- License
- Open Source (Proof-of-Concept)
- Released
- 2025-10-27
- Model Variants
- 17M, 32M parameters
- Base Architecture
- ColBERT late-interaction
- Primary Use Case
- Information Retrieval
- 17M and 32M parameter variants
- Designed for local edge device deployment
- Open-source proof-of-concept for researchers
Overview
mxbai-edge-colbert-v0 is a family of extremely efficient ColBERT models developed by Mixedbread for late-interaction research. These models are designed for developers and researchers who require high-performance retrieval that can run locally on edge devices or at scale in the cloud. They represent a significant step forward in long-context tasks with unprecedented efficiency. The models are available as open-source proof-of-concepts for the developer community to serve as a foundation for future experiments.
Efficient Retrieval Architecture
- Small Footprint: The model family includes extremely efficient versions with 17 million and 32 million parameters.
- Edge Deployment: Designed for high-performance retrieval that can run locally on edge devices or at scale in the cloud.
- Open Source Research: The models are released as open-source proof-of-concepts for the developer and research community.
Search and Retrieval Performance
- Long-Context Support: Represents a significant step forward in long-context tasks with unprecedented efficiency.
- Research Foundation: Serves as a solid foundation backbone for future experiments in late-interaction models.
Benchmarks
- Outperforms ColBERTv2 on BEIR benchmarks
Use cases
- Researchers use the 17M and 32M models to conduct experiments on late-interaction retrieval efficiency.
- Developers deploy the models locally on edge devices to perform high-performance search without cloud dependency.
Specifications
- Parameters
- 17M and 32M versions
- License
- Open Source (Proof-of-Concept)
- Released
- 2025-10-27
- Model Variants
- 17M, 32M parameters
- Base Architecture
- ColBERT late-interaction
- Primary Use Case
- Information Retrieval
- Deployment Targets
- Cloud and Edge Devices
- Benchmark Target
- BEIR (short-text)
- Release Status
- Open Source Proof-of-Concept
How it works
- Model Distillation: Mixedbread conducts experiments to improve retrieval and late-interaction models, distilling them into smaller parameter counts.
- Edge Optimization: The models are optimized for efficiency to allow them to run on local devices or at scale in the cloud.
- Community Access: Developers and researchers access the models as open-source proof-of-concepts for their own retrieval tasks.
FAQ
What are the parameter counts for mxbai-edge-colbert-v0?
The family includes two versions with 17 million and 32 million parameters respectively.
Can these models run on edge devices?
The models are designed to support retrieval at all scales, from large-scale cloud deployments to local execution on any device.
Is the model open source?
Mixedbread released these models as open-source proof-of-concepts to serve as a foundation for future experiments in late-interaction research.
Added by AntSpace from public sources (mixedbread.com).