Granite 4.1 8B

Granite 4.1 8B – Dense Decoder-Only Enterprise Language Model

Granite 4.1 8B is IBM's balanced-tier open-source language model, part of the Granite 4.1 family (3B, 8B, 30B). It uses a dense, decoder-only transformer architecture — deliberately avoiding sparse MoE routing — to deliver predictable latency and stable token usage for enterprise workloads. IBM documents the 8B instruct variant as consistently matching or outperforming the previous-generation Granite 4.0-H-Small, a 32B MoE model with 9B active parameters.

Architecture

Granite-4.1-8B is built on a dense decoder-only transformer with: GQA (Grouped Query Attention), RoPE positional embeddings, MLP with SwiGLU activation, RMSNorm, and shared input/output embeddings. No sparse routing, no MoE layers.