Snowflake Partners with AI21's Jamba-Instruct to Enable Enterprises in Analyzing Long Documents

Snowflake, the leading data cloud provider, has officially integrated AI21 Labs’ Jamba-Instruct LLM into its Cortex AI service. This new addition is designed to help Snowflake's enterprise clients develop generative AI applications—such as chatbots and summarization tools—that efficiently handle lengthy documents without sacrificing quality or accuracy.

Starting today, the Jamba-Instruct model empowers organizations to leverage large files—a necessity in many enterprises. While AI21 Labs is a significant partner, Snowflake collaborates with various LLMs to enhance its generative AI ecosystem. Recently, Snowflake partnered with Meta to incorporate the Llama 3.1 LLM family and launched its proprietary model, 'Arctic,' reflecting its aggressive strides in the generative AI space.

Benefits of Jamba-Instruct for Snowflake Users

In March, AI21 Labs unveiled Jamba, an open generative AI model that combines a transformer architecture with a novel, memory-efficient Structured State Space model (SSM). Jamba stands out by offering a remarkable 256K context window, leading to a threefold increase in throughput for long contexts compared to similar models. This efficiency made way for Jamba-Instruct, an instruction-tuned version that includes advanced training, chat capabilities, and safety features for enterprise applications.

Launched on AI21’s platform in May, Jamba-Instruct is now part of Cortex AI, Snowflake's no-code, fully managed service for building powerful generative AI applications. “With its large context window, Jamba-Instruct processes up to 256K tokens—about 800 pages of text—making it an invaluable tool for extensive document management,” said Baris Gultekin, Snowflake's head of AI.

For example, financial analysts can utilize Q&A tools to extract insights from lengthy 10-K filings, while clinicians can rapidly analyze extensive patient reports for relevant data. Retailers can also create chatbots capable of maintaining coherent, reference-based dialogues with customers.

Gultekin emphasized that the model's extensive context window simplifies the creation of retrieval-augmented generation (RAG) pipelines, allowing for efficient information retrieval and supporting guided prompts for specific tones during content generation.

Cost Efficiency

In addition to its long-document capabilities, Jamba-Instruct offers significant cost savings for Snowflake customers. The model’s hybrid design and mixture-of-experts (MoE) technology make its expansive context window more economically accessible compared to other instruction-tuned transformer models. Coupled with Cortex AI’s serverless inference and consumption-based pricing model, enterprises only pay for the resources they use, eliminating the need for costly dedicated infrastructure.

“Organizations can effectively balance performance, cost, and latency by harnessing Snowflake’s scalability alongside Jamba-Instruct’s efficiency. Cortex AI’s architecture enables seamless scaling of compute resources,” explained Pankaj Dugar, SVP & GM for North America at AI21 Labs.

Currently, Cortex AI supports a variety of LLMs, including Snowflake’s Arctic model and offerings from Google, Meta, Mistral AI, and Reka AI. “We strive to give our customers the flexibility to choose between open-source and commercial models, addressing their specific needs without complicating data governance,” Gultekin added.

The selection of models is expected to grow, with new options—especially from AI21—set to arrive in the coming months. Gultekin highlighted that customer feedback plays a crucial role in evaluating and integrating LLMs to ensure the right tools are available for various use cases, including automated business intelligence, conversational assistants, and text summarization.

Snowflake recently acquired TruEra to help clients navigate the expanding landscape of model choices. Gultekin noted that TruEra’s TruLens allows users to experiment with LLMs and assess the best fits for their needs.

Today, over 5,000 enterprises utilize Snowflake’s AI capabilities, focusing on key applications like automated BI, conversational assistants, and text summarization.