IBM has introduced Granite 3.0, the latest version of its large language models (LLMs) designed for enterprise use. These models combine performance with cost-efficiency, offering businesses a reliable tool for various applications, from cybersecurity to text generation.
“Granite 3.0 8B Instruct is a robust model that can be customized to meet enterprise needs,” IBM said in a blog post. “It was developed with over 12 trillion tokens and supports 12 natural languages and 116 programming languages, making it versatile for multiple industries.
The new release also highlights IBM’s commitment to open-source development. All Granite models are available under the Apache 2.0 license.
“We want to foster transparency and trust in AI,” IBM added, referencing its decision to disclose training data and methodologies.
Granite 3.0 models are now available on platforms like IBM’s watsonx, Google Vertex AI, and Nvidia, offering easy access for businesses. Updates later this year will improve language support and add features like image-based input.
IBM continues its push for sustainable tech with Granite 3.0 being trained on Blue Vela, a system powered entirely by renewable energy. The models have been optimized for various industries, excelling in fields like cybersecurity and code generation.