IBM Launches Granite 3.0 AI Models, Reinforcing Its Commitment to Open Source

23.10.2024

IBM introduced its most advanced AI model family, Granite 3.0, at the annual TechXchange event. The Granite 3.0 lineup includes general-purpose language models (8B and 2B Instruct/Base configurations), safety-focused Guardian models, and a Mixture-of-Experts series for various applications.

IBM claims these models can compete with leading providers in academic and industry benchmarks, excelling in tasks like Retrieval Augmented Generation (RAG), classification, summarization, and entity extraction.

A key feature of Granite 3.0 is IBM’s commitment to open-source AI, with models released under the Apache 2.0 license. This allows businesses to achieve task-specific performance at a lower cost by combining the models with proprietary data.

IBM emphasizes transparency and safety, publishing a technical report and responsible use guide for Granite 3.0. The 8B Instruct model has outperformed similar open-source models from Meta and Mistral and leads in safety metrics on IBM’s AttaQ benchmark.

The Granite Guardian 3.0 models enhance safety by monitoring user prompts and LLM responses for risks. The entire suite is available for download on HuggingFace, with commercial options on IBM’s watsonx platform. IBM is also focusing on developing advanced AI agent technologies for greater autonomy and problem-solving capabilities, with new features planned for 2025.