IBM Releases Granite 4.1 family of models (research.ibm.com)

IBM has launched the Granite 4.1 family of models, a significant enhancement aimed at streamlining enterprise AI applications. This release includes small language models (SLMs) alongside cutting-edge speech, vision, and embedding models, tailored for real-world AI tasks. Notably, the Granite 4.1 language models boast improvements in instruction following and tool calling, with new architectures allowing impressive performance at reduced operational costs. IBM's meticulous training approach focused on high-quality data over sheer volume, with models capable of processing up to 512K tokens, making them adept at handling lengthy documents. The Granite 4.1 suite also features advanced models for Vision and Speech, capable of high accuracy in tasks such as table extraction and multilingual speech recognition. The Granite Guardian model enhances safety and quality by monitoring AI outputs for harmful content, reflecting a growing trend in the AI community to integrate risk assessment directly within AI systems. The comprehensive nature of this release emphasizes modularity and efficiency, providing enterprises with a robust framework for diverse AI applications. All models are released under an open-source Apache 2.0 license, reinforcing IBM's commitment to transparent innovation in AI.