IBM's new Granite 4.2 models ride the wave of interest in local LLMs
Summary
IBM has introduced the Granite 4.2 series of large language models (LLMs) designed for users to download and run on their own hardware. These models come in three sizes and focus on improved reasoning abilities, handling long texts, and supporting tools for tasks like web searching.Key Facts
- Granite 4.2 models are available in 3 billion, 8 billion, and 30 billion parameter versions, with parameters being settings that shape the AI's capabilities.
- All versions use a decoder-only design, a common type of AI language model architecture.
- The models can handle very long inputs of up to 128,000 tokens (words or parts of words) at once.
- The 8B and 30B models underwent extra training to better use external tools like terminals, web search, and software utilities.
- The 3B model supports tools but with less specialized training than the larger versions.
- The main feature of this release is enhanced "reasoning" abilities, meaning the models can follow step-by-step logic to give more accurate answers.
- These reasoning improvements may cause slower responses and require more computing power.
- IBM focuses on reliability and predictable deployment rather than speed or cutting-edge innovation compared to competitors.
- Interest in locally hosted models like Granite is growing because they reduce reliance on expensive cloud-based AI services and avoid ongoing fees.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
