IBM发布Granite4.2包含3B至30B参数支持原生推理
IBM推出Granite 4.2开源大语言与语音模型。该系列涵盖3B、8B及30B三种参数规格。模型专为企业级代理式人工智能(Agentic AI)工作流打造。
Granite 4.2具备原生逐步推理(Reasoning)能力。模型可自主评估工具选择顺序,精确处理多步骤指令。软件工程代理可借此在终端环境中完成代码开发。
模型支持云端、私有云及边缘端部署。小尺寸模型可高效处理高吞吐量任务。大尺寸模型能支持深度推理与复杂编程场景。
IBM推出Granite4.2模型
模型具备原生逐步推理能力
支持云端及边缘端灵活部署
IBM Launches Granite 42 Models 3B To 30B Parameters Native Reasoning Enterprise Agents
IBM has officially released its updated Granite 4.2 suite of open language and speech models to advance enterprise agentic Artificial Intelligence (AI). Available across 3B, 8B, and 30B parameter sizes, these dense architecture models are purpose-built for autonomous enterprise workflows. The update introduces native step-by-step reasoning capabilities, which allow AI agents to independently formulate multi-step plans, evaluate trade-offs prior to decision-making, and self-correct potential operational errors before execution.
Granite 4.2 features enhanced tool-calling capabilities that enable models to evaluate external software applications and determine precise execution sequences rather than running blindly. In software development contexts, specialized agents built on Granite 4.2 can navigate complex codebases, manage multi-stage engineering tasks, and operate autonomously inside terminal environments. The flexible parameter architecture provides organizations with deployment options across cloud, on-premises, and edge infrastructure, enabling small models to manage high-throughput agentic tasks while reserving larger variants for complex reasoning.
The release reflects an ongoing paradigm shift from reactive conversational chatbots to task-oriented autonomous agents capable of handling ambiguous business requests. By embedding native step-by-step reasoning into open enterprise models, IBM provides scalable infrastructure for reliable workflow automation. This approach addresses key corporate requirements for safety, precision, and tool integration across diverse enterprise deployment environments.
Key Takeaways:
IBM launched Granite 4.2 models in 3B, 8B, and 30B parameter configurations. Source: Original Article
Native reasoning enables autonomous agents to plan, execute, and self-correct complex workflows. Source: Original Article
Granite 4.2 supports deployment across cloud, on-premises, and edge computing environments. Source: Original Article
查看原文 →
View Original →