LLM ReferenceLLM Reference

Granite 4.0 H Small

granite-4.0-h-small

Open Source

About

IBM Granite 4.0 H Small is the flagship hybrid MoE model with 32B total parameters and 9B active parameters. Architecture: 4 attention + 36 Mamba2 layers, 72 total experts with 10 active. Highest-performance Granite 4.0 text model. Benchmarks: MMLU 78.44, HumanEval 88%, GSM8K 87.27, IFEval 87.55, SALAD-Bench 97.3. Artificial Analysis Intelligence Index 23. Apache 2.0.

Granite 4.0 H Small has a 128K-token context window.

Capabilities

VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode Execution

Rankings

Specifications

FamilyGranite 4
Released2025-10-02
Parameters32B total / 9B active
Context128K
ArchitectureHybrid MoE: 4 attention + 36 Mamba2 layers, 72 experts / 10 active, NoPE position embedding

Created by

Creating reliable and adaptable AI solutions

Armonk, New York, United States
Founded 1945
Website