Jamba-Instruct Launches on Amazon Bedrock

AI21 Labs' latest foundation model, Jamba-Instruct, has been added to Amazon Bedrock. This model features a massive 256K context window and competitive pricing, making it optimized for long document summarization, question answering, and text generation tasks.

Jamba-Instruct is designed based on the world's first production-grade SSM-Transformer hybrid architecture. By combining the existing Transformer structure with SSM architecture, it secures performance, quality, and cost efficiency all at once.

Key features are as follows:

• 256K Context Window: Strengthens the retrieval stage of RAG pipelines, supports long document summarization and multi-document synthesis
• Cost-Efficient Processing: Reduces costs by processing long contexts with fewer cloud resources through the hybrid architecture
• Best Cost-Performance: Maintains high price competitiveness while delivering performance that matches or exceeds competing models on major benchmarks

Through Amazon Bedrock, enterprises can leverage the model in a reliable and scalable AWS environment. As a fully managed service, it provides robust security features such as VPC configuration, encryption, and compliance monitoring without the burden of infrastructure maintenance.