Granite

8 MLX aliases · IBM Granite 4.2 (3B / 8B / 30B) + Granite 4 hybrid state-space small models.

Pick one

One command per line of this family, smallest download first. Your Mac needs the download size in free memory plus room for macOS and the context window; the hardware tiers page has the engine's picks for every RAM size. The first run downloads the weights and starts an OpenAI-compatible server on http://localhost:8000/v1.

• rapid-mlx serve granite4-h-micro-4bit
  Granite 4 · 1.8 GB download
• rapid-mlx serve granite-4.2-3b-4bit
  Granite 4.2 · 2.07 GB download

IBM's Granite line — the Granite 4.2 dense models in 3B, 8B and 30B, published by IBM as 4-bit and 8-bit MLX checkpoints, plus the Granite 4 hybrid state-space + attention small models for low-RAM Macs.

family

Granite

aliases

8

lines

2

install

rapid-mlx serve <alias>

OpenAI base URL

http://localhost:8000/v1

Download

Every alias on this page downloads with one command — the pull buttons in the tables below copy it. All 8 aliases on this page are mirrored on the rapid-mlx CDN — with automatic mid-pull fallback to Hugging Face if a mirror file slows down. Weights land in the standard Hugging Face cache, and rapid-mlx serve pulls automatically on first use. Live mirror status →

Lines in this family

• Granite 4.2 (6 aliases)
• Granite 4 (2 aliases)

Granite 4.2 · 6 aliases

Dense 3B, 8B and 30B in IBM's own 4-bit and 8-bit MLX builds. Tool calls use the qwen3_coder_xml parser and reasoning the qwen3 parser.

parser: qwen3_coder_xml

Granite 4 · 2 aliases

H-Micro + Tiny variants. Both run hybrid cache — long-context cost bounded by SSM step, not by KV cache size.

parser: hermes

Notes & caveats

• granite4-h-micro-4bit is sub-2 GB on disk — runs anywhere.
• Hybrid cache enabled — long-context cost bounded by SSM step, not by KV cache size.

Where next