IBM releases Granite 4.2 local models with enhanced agent capabilities
Ars Technica AI · rss · 2026-08-26
IBM has launched the Granite 4.2 family of open-weight LLMs available in 3B, 8B, and 30B parameter variants, featuring a native 128K context window. The 8B and 30B models have undergone "agentic reinforcement learning" training, equipping them with expanded capabilities such as terminal usage, web searching, and external tool use. While the 3B model also supports tools, it lacks this specialized training. Designed for download and self-hosting, these models cater to the growing interest in local LLMs.
More from coding & agent
- Testing Copperkit App for CNC PCB Milling — debreuil · 2026-08-27
- MCP Tool Design Anti-pattern: Avoid 1:1 Mapping to REST API — arpit_bhayani · 2026-08-27
- AgentSafeFS Hits npm: A Propose-Commit Safety Layer for AI Agents' File Writes — tomaszteee · 2026-08-27
- Opinion: The Real Unlock is Making 3D Generation an On-Demand Tool for Agents — eyishazyer · 2026-08-27
- Antigravity's /learn Turns Successful Coding Trajectories Into Reusable Skills — rseroter · 2026-08-27
- Opinion: "Looks Better" is Not a Valid Prompt Validation Method — bgoncalves · 2026-08-27