Granite 3.2 introduces multi-modal and reasoning AI models, enhancing enterprise efficiency and scalability.
IBM has unveiled the next generation of its Granite large language model (LLM) family, Granite 3.2, designed to deliver small, efficient, and practical AI solutions for enterprises. The new models offer reasoning, vision, and guardrail capabilities, all available under the permissive Apache 2.0 license on Hugging Face and other platforms.
Granite 3.2 includes a new vision language model (VLM) for document understanding tasks, matching or exceeding the performance of larger models on key benchmarks. Enhanced reasoning capabilities in the 3.2 2B and 8B models allow for optimized efficiency, with the ability to switch reasoning on or off as needed. Additionally, the Granite Guardian safety models have been slimmed down by 30% while maintaining performance.
“The next era of AI is about efficiency, integration, and real-world impact – where enterprises can achieve powerful outcomes without excessive spend on compute,”
Sriram Raghavan, VP, IBM AI Research.
IBM’s strategy focuses on delivering smaller, specialized AI models that are cost-effective and scalable. The Granite 3.1 8B model recently achieved high marks on accuracy in the Salesforce LLM Benchmark for CRM, demonstrating the efficacy of this approach.
David Tan, CTO of CrushBank, praised the new models, stating, “Granite 3.2 takes it further with new reasoning capabilities, and we’re excited to explore them in building new agentic solutions.”
Alongside Granite 3.2, IBM is releasing the next generation of its TinyTimeMixers (TTM) models, which offer long-term forecasting capabilities for various industries. These developments mark another step forward in making AI more accessible and valuable for modern enterprises.