Box and IBM partner to bring Llama models to Box AI

Today, Box and IBM are expanding their strategic partnership by integrating Box AI with IBM’s watsonx. This brings trusted enterprise AI, including IBM’s open-source Granite models and Meta's powerful open-weight Llama models, directly into Box AI. Enterprises face the critical challenge of harnessing AI's power securely across vast stores of unstructured content. This integration also incorporates IBM watsonx.governance, providing robust tools for responsible AI deployment.

Llama models demonstrate strong enterprise performance

Following a successful preview period where over 100,000 users leveraged Meta's Llama models within Box AI, we will soon be expanding the availability of the Llama model family to all Box customers. Our internal evaluations demonstrate that Llama models perform strongly on demanding enterprise use cases, particularly excel at dealing with complex information. The Llama model family will be the first open-weight models supported by Box AI and we are exciting to bring their flexibility, transparency, and effectiveness to enterprise content. These models can combine strong performance, with greater visibility into model architecture and operation. Enterprises can deeply understand the model that is supporting their AI experiences, while not compromising on accuracy or efficiency.

Llama 4 achieves significant gains over Llama 3

The Llama family also showcases substantial improvement over time, with new releases increasing performance and efficiency. Llama 4 Maverick, for instance, demonstrated a 33% gain in accuracy over Llama 3 Nemotron in our evaluations. This significant leap highlights the rapid evolution and increasing power of the Llama family, offering businesses enhanced performance and efficiency for tackling their most challenging content-related tasks. The adoption by over 100,000 preview users further validates the real-world value these models bring.  

A key finding from our Box AI Enterprise Eval is Llama 4's exceptional ability to handle information complexity within business documents. The models achieved near-perfect accuracy extracting straightforward contract details. Critically, Llama 4 Scout reached 92% accuracy when extracting information requiring the interpretation of complex logic – a vital capability for reliably automating nuanced tasks like detailed contract analysis or compliance reviews. This high degree of accuracy on complex tasks signifies Llama 4's readiness for mission-critical enterprise workflows.

Box AI now integrates with IBM watsonx

Box AI with IBM watsonx, streamlines workflows by accelerating content research, creation, and review processes, allowing teams to focus on strategic work and impactful decision-making.

IBM has enabled Box AI with IBM watsonx to elevate employee workflows, enabling seamless document and content interaction. This implementation demonstrates how Box effectively delivers AI capabilities directly into business environments, creating value across various industries.

Additionally, Box leverages IBM watsonx.governance for lifecycle management of AI models, implementing monitoring and guardrails. By using this technology internally, Box ensures responsible AI deployment throughout the entire model lifecycle. This provides Box customers with tools to manage risk, compliance, and AI lifecycle processes, while maintaining auditable and trustworthy AI-driven insights.

Putting this partnership to work across your organization

Earlier this month, we evaluated Llama 4 and its predecessor Llama 3, we found that Llama 4 models offer key advantages for enterprises using Box AI:

• Reliable accuracy for complex tasks: The demonstrated 92% accuracy on complex logic extraction enables dependable automation and insight generation.  
• Handles nuance: Moves beyond basic keyword matching to understand intricate requirements common in legal, financial, and technical documents.  
• Measurable performance improvement: The 33% gain over Llama 3 translates to tangible efficiency and more capable AI applications.  

And that impact can be felt across a variety of industries:

• Financial services firms can accelerate fraud detection and risk analysis by extracting patterns from large volumes of transaction data. They are also able to enable real-time anomaly detection while ensuring compliance and auditability.
• Healthcare & life sciences companies canstreamline clinical trials by leveraging AI to analyze patient records, clinical notes, and study results. Teams can also identify candidate matches faster while maintaining governance controls to ensure data integrity.
• Law firms can simplify contract analysis with advanced data extraction techniques and facilitate quicker reviews of key clauses across extensive legal documents backed by a transparent AI process.
• Government agencies can enhance public sector efficiency through automated document processing for permits and compliance reports. They are also able to reduce administrative burdens while ensuring accuracy with built-in governance measures.
• Insurance firms can expedite claims processing by extracting critical details from accident reports and policy documents. They may also leverage AI to improve turnaround times while enhancing accuracy in regulatory compliance efforts.

Get started today

Box AI with IBM watsonx, includingMeta Llama models, is now available to Box Enterprise Advanced customers through Box AI Studio and APIs. IBM is also an authorized reseller, making it easy for customers to access IBM’s flagship family of business-ready AI models alongside Box’s intelligent content management capabilities. Begin transforming your content workflows with enhanced accuracy and efficiency today.