With $2 billion in generative AI business, IBM expands its enterprise AI capabilities with cutting-edge Granite 3.0 models.

IBM is taking a bold step forward in the enterprise AI space with the debut of Granite 3.0, its third generation of large language models (LLMs). These open-source models are designed to cater to the growing demand for AI solutions in the enterprise sector, where IBM already boasts a $2 billion generative AI business.

Granite 3.0 offers both general-purpose models, including 2 billion and 8 billion parameter versions, and specialized Mixture-of-Experts (MoE) models tailored for various business applications. Additionally, IBM has introduced Granite Guardian 3.0 models, which emphasize safety and optimized guardrails to ensure responsible AI use. The new models are accessible through IBM’s watsonX platform, Amazon Bedrock, Amazon Sagemaker, and Hugging Face.

A key innovation behind Granite 3.0 is IBM’s focus on fine-tuning models for specific enterprise needs using its InstructLab capabilities. These models are expected to support industries ranging from customer service and IT automation to cybersecurity and business process outsourcing (BPO).

The models were trained using 12 trillion tokens of multilingual and code data, curated by IBM’s data model factory team. IBM claims the new generation surpasses competitors like Google and Anthropic in performance, while also balancing size and inference costs, making them ideal for enterprise-scale applications.

IBM’s emphasis on safety and trust, with advanced “Guardian” models, ensures these LLMs are equipped to prevent harmful outputs. As IBM continues to push the boundaries of enterprise AI, the Granite 3.0 models represent a leap forward in both innovation and practicality.