ChipsThe New Stackabout 2 hours ago
IBM launches Granite 4.2 reasoning models
4 min read
IBM launched Granite 4.2, a family of open-weight, dense, decoder-only reasoning LLMs pre-trained from scratch in 3B, 8B, and 30B parameter sizes. The models contrast with hybrid Mamba/attention architectures like Nvidia's Nemotron 3, following IBM's earlier Granite 4.0 experiments with dense, dense-hybrid, and hybrid MoE variants.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyIBM
Summary by ByteBrief
