Jamba is a family of open large language models developed by AI21, built on a hybrid SSM-Transformer architecture with Mixture-of-Experts (MoE) layers. Designed specifically for enterprise deployment, Jamba models offer state-of-the-art performance in quality, speed, and long-context processing, supporting up to 256,000 tokens. The models are available in "Large" and "Mini" versions, both optimized for business-critical tasks such as retrieval-augmented generation (RAG), grounded question answering, and structured output. Jamba can be deployed flexibly—via SaaS, self-hosted on-premises, or through cloud partners—enabling organizations to maintain full control over their data and meet stringent security requirements. Released under a permissive open model license, Jamba is suitable for research and commercial use, making it a leading choice for enterprises seeking high-quality, efficient, and secure AI solutions.

Key Features

Hybrid SSM-Transformer-MoE Architecture Combines Transformer and Mamba layers with Mixture-of-Experts for optimal performance.

• Enables high throughput and small memory footprint
• Provides configurable architecture for resource and use-case optimization

Long Context Window (256K tokens) Handles very large documents and multi-document workflows.

• Excels at RAG and long-context question answering
• Supports summarization and analysis of lengthy texts

Enterprise-Grade Performance Outperforms leading open and closed models in quality and speed.

• Superior benchmark results (Arena Hard, CRAG, LongBench)
• Up to 2.5X faster inference than comparable models

Flexible, Secure Deployment Supports SaaS, VPC, on-premises, and cloud partner integrations.

• Full data control for regulated industries
• Available as NVIDIA NIM microservice and via platforms like Vertex AI and Azure AI

Advanced Business Capabilities Optimized for enterprise tasks and developer features.

• Function calling and structured JSON output
• Batch API for high-volume, asynchronous processing

Multi-language Support Supports major European and Middle Eastern languages.

• English, Spanish, French, Portuguese, Italian, Dutch, German, Arabic, Hebrew

Benefits

Data Security and Privacy Keeps sensitive information within organizational boundaries.

• Enables private, on-premise, or VPC deployments
• Meets compliance needs for finance, healthcare, and legal sectors

Operational Efficiency Reduces latency and manual workload in enterprise workflows.

• Low-latency responses for real-time applications
• Batch processing for handling large data spikes

Scalability and Cost Efficiency Handles high-volume, complex tasks at competitive cost.

• Efficient hardware utilization
• Cost-effective for large-scale document processing and analysis

Customizability and Alignment Allows organizations to shape model behavior.

• Supports alignment pipelines for safety and policy compliance
• Customizable outputs to fit enterprise standards