As part of our longstanding partnership with Amazon Bedrock, we are pleased to announce the addition of our latest and most advanced Foundation Model, Jamba-Instruct, to AI21 Labs’ offering on Amazon Bedrock. Due to its 256K context window and competitive pricing, Jamba-Instruct is the economic, the company said on 26 June 2024.
Jamba-Instruct: Built from hybrid SSM-Transformer architecture As the world’s first production-grade SSM-Transformer model, Jamba-Instruct is built from the groundbreaking Jamba architecture, a novel blocks-and-layers approach that successfully integrates SSM architecture with the conventional Transformer architecture. Jamba-Instruct leverages its hybrid architecture to achieve leading performance, quality, and cost efficiency. As a result, it is ideal for enterprises who wish to build and scale reliable GenAI workflows, without running into the common barrier of prohibitive compute costs. Jamba-Instruct stands out for its: Impressive context window length : Offering a 256K context window, Jamba-Instruct supports a range of GenAI enterprise use cases, such as enhancing the retrieval stage of a RAG pipeline, summarizing lengthy documents, and offering more comprehensive synthesis over multiple documents. Cost-efficient processing: Due to its novel hybrid architecture, Jamba-Instruct handles lengthy context on a smaller cloud footprint than ever before, driving down costs.
Cost-efficient processing: Due to its novel hybrid architecture, Jamba-Instruct handles lengthy context on a smaller cloud footprint than ever before, driving down costs. Best value per cost: Surpassing or rivaling competitors across common quality benchmarks, Jamba-Instruct’s highly competitive pricing gives enterprises the best value for their money.
For enterprise buyers, the useful next step is to map the claim to controls, data residency and measurable outcomes—and to ask for evidence on failure modes, oversight and auditability before treating the release as production-ready fact.