IBM has launched the latest iteration of its artificial intelligence models, the Granite 3.0, tailored for business applications.

The models, designed to leverage the growing trend of generative AI in enterprises, will be open-source, providing an alternative to competitors like Microsoft.

IBM’s new flagship language models are reported to either “outperform or match” the capabilities of other leading models on various benchmarks.

The Granite 3.0 models are set to be commercially available on IBM’s Watsonx platform, which offers a paid service to run these models in data centres post-customisation.

Additionally, select models from the new Granite series will be accessible through Nvidia’s software tools, facilitating AI integration for businesses.

IBM research director Dario Gil was quoted by Reuters as saying that the models were trained using Nvidia’s H100 GPUs, highlighting the collaboration between the two tech giants.

The Granite 3.0 family encompasses a range of models, including General Purpose/Language, Guardrails & Safety, and Mixture-of-Experts.

These models, such as the Granite 3.0 8B and 2B language models, are engineered as ‘workhorse’ models for enterprise AI.

They deliver robust performance across various tasks such as retrieval augmented generation, classification, summarisation, entity extraction, and tool use. The models are compact and versatile, designed for fine-tuning with enterprise data and integration into diverse business workflows.

The latest development comes after IBM announced the acquisition of Prescinto, an Indian company specialising in asset performance management software-as-a-service for the renewable energy sector.