Granite 4.2 brings native reasoning to enterprise agents (research.ibm.com)

IBM has unveiled the Granite 4.2 language models, a significant upgrade that introduces native reasoning capabilities to enterprise agents, allowing them to handle complex workflows with autonomy. With models available in 3B, 8B, and 30B parameter sizes, Granite 4.2 enhances AI’s ability to follow intricate instructions, make informed decisions, and correct errors proactively. This evolution is critical as enterprises increasingly demand that AI systems not only generate responses but also navigate multifaceted tasks that require planning, executing, and adapting in real-time. The Granite 4.2's innovative training methodology, which incorporates multi-stage reinforcement learning and the use of synthetic code from IBM's CodeAlchemy, equips these models for advanced reasoning in software engineering and coding tasks. Additionally, IBM launched two new speech models, Granite Speech 5.0 Turbo CTC, optimized for edge devices. These models excel at automatic speech recognition (ASR), achieving unprecedented processing speeds, making them suitable for high-volume transcription tasks. Overall, these advancements demonstrate IBM’s commitment to building AI systems that can effectively operate in real-world scenarios, transforming the landscape for enterprise applications.