Dive Brief:
- IBM is expanding its AI menu with the launch of Granite 3.2, a family of small AI models designed for enterprise use, the vendor said Wednesday.
- Granite 3.2 Instruct is aimed at completing complex reasoning tasks, mathematical problems and general language requests. Enterprise customers can switch off chain-of-thought reasoning capabilities, which are often more expensive and time-consuming, within the Granite 3.2 2B and 8B Instruct models to optimize compute efficiency, the vendor said.
- IBM also added an assortment of models with varying context lengths and forecasting horizons, including a smaller option of Granite Guardian for risk assessment. The latest additions come a few months after the third generation of the Granite series launched in October. “Much of our ongoing research aims to take advantage of the inherently longer, more robust thought process of Granite 3.2 for further model optimization,” IBM said in the release.
Dive Insight:
Large language models might have dominated the enterprise generative AI conversation initially, but business leaders are turning to lightweight versions as they look to rein in costs and boost efficiency.
Smaller models typically use less computing power and are often tailored to complete specific tasks. Enterprises have struggled in both areas with LLMs resulting in delayed projects due to computing availability and prioritization of domain-specific capabilities in AI purchases.
Vendor options are plentiful, from Google’s lightweight Gemma models to Microsoft’s Phi family and OpenAI’s o3-mini.
“We have been very vocal for about a year that smaller models and more reasonable training times are going to be essential for enterprise deployment of large language models,” IBM CEO Arvind Krishna said during IBM’s Q4 2024 earnings call in January. “We see as much as 30 times reduction in inference costs using these approaches.”
The company attributed around $1.5 billion in new bookings during the latest quarter to its generative AI business. Retail and commercial banking company NatWest and defense manufacturer Lockheed Martin are among the enterprise clients already utilizing IBM’s Granite models, according to Krishna.
All Granite 3.2 models are available under the Apache 2.0 license on Hugging Face. Customers can also access select models on IBM watsonx.ai, LM Studio, Ollama and Replicate.
“The release of Granite 3.2 marks only the beginning of IBM’s explorations into reasoning capabilities for enterprise models,” IBM said in a release.
