AI21's Jamba-Instruct Integrated into Snowflake Cortex

Through a partnership with Snowflake, AI21 is integrating its latest foundation model, Jamba-Instruct, into Snowflake Cortex. Snowflake Cortex is a fully managed service that allows customers to analyze text data and build AI applications using a variety of LLMs and vector search capabilities.

Integration through Snowflake Cortex offers the following benefits:

• Seamless deployment of LLMs: Natively integrated with the Snowflake Data Cloud, simplifying the connection between data and GenAI models.
• Secure data access: Data remains within Snowflake's security perimeter, allowing full use of existing security, scalability, and governance capabilities.
• Infrastructure management: Optimizes the complex GPU infrastructure required to run and scale GenAI solutions.

Jamba-Instruct is based on a hybrid architecture that combines the existing Transformer structure with the new SSM structure. In particular, it offers a long context window of 256K, delivering superior performance in handling long contexts compared to competing models such as Mixtral 8x7B.

This long-context capability is highly advantageous for automated terms-and-conditions generation in the financial sector, customer-facing chatbots that need to maintain long conversations, and building advanced RAG engines that leverage vast enterprise knowledge bases. In addition, competitive pricing supports cost-efficient construction of large-scale GenAI applications.