gemma-4-26b-4bit
Run Gemma 4 26B (A4B MoE) on a Mac
Mixture-of-experts done right for local use: 26 B of knowledge, but only ~4 B active per token — so it decodes at 78.5 tok/s on a 48 GB M4 Pro, far faster than the dense 27B class.
MoE · 26 B total / 4 B active · 4-bit32 GB Macs · fast high-quality chat · Tier-1 family
Google 256K context Base model intelligence 26.1 AA
One command
rapid-mlx serve gemma-4-26b-4bit
Weights (13 GB) download on first run; you get an OpenAI-compatible
endpoint at http://localhost:8000/v1. No rapid-mlx yet? It's one line —
curl -fsSL https://rapidmlx.com/install.sh | bash — or take the
desktop app.
Point a client at it. OpenAI SDK: base_url="http://localhost:8000/v1",
model="gemma-4-26b-4bit". Claude Code:
ANTHROPIC_BASE_URL=http://localhost:8000 (no /v1 suffix) —
see the Claude Code guide.
Measured on real hardware
Headline machine · Mac mini M4 Pro · 48 GB
All measured machines
| Machine | Decode | First token | Peak memory | Cold boot | Weights |
|---|---|---|---|---|---|
| Mac mini M4 Pro · 48 GB rapid-mlx 0.15.4 · 2026-10-02 raw data | 78.5 tok/s | 3.1 s | 20.9 GB | — | 13 GB |
Each row cites its own run; numbers are never copied across machines. Mac mini M4 Pro · 48 GB: rapid-mlx 0.15.4, 2026-10-02 — raw data + method — 4 concurrent streams: 121.6 tok/s aggregate.
Community runs, one page per Mac: Apple M4 Max · Apple M4 Pro.
Will it fit your Mac?
Measured 20.9 GB Metal peak (16.0 GB process RSS) on a 48 GB M4 Pro across the whole benchmark run — up to a ~16K-token prompt and four concurrent streams, including the engine's prefix cache, which defaults to 20% of RAM. A short chat needs less; long contexts grow the KV cache, so leave headroom.
For the conservative install-default placement see the hardware tiers table; to compare against every model your RAM can hold, use the live picker.
Variants & profiles
gemma-4-26b-qat-4bit— The QAT 4-bit variant — quantisation-aware weightsgemma-4-12b-4bit— Gemma 4 12B dense — smaller, takes images too
FAQ
How much memory does Gemma 4 26B (A4B MoE) need on a Mac?
Measured 20.9 GB Metal peak (16.0 GB process RSS) on a 48 GB M4 Pro across the whole benchmark run — up to a ~16K-token prompt and four concurrent streams, including the engine's prefix cache, which defaults to 20% of RAM. A short chat needs less; long contexts grow the KV cache, so leave headroom.
How fast is Gemma 4 26B (A4B MoE) on Apple Silicon?
We measured 78.5 tok/s on a Mac mini M4 Pro · 48 GB, rapid-mlx 0.15.4, 2026-10-02; peak memory 20.9 GB on that run. Every row on this page keeps its own chip, version and date — no number is copied across machines.
How do I run Gemma 4 26B (A4B MoE) locally?
rapid-mlx serve gemma-4-26b-4bit — the weights download on first run and you get an OpenAI-compatible endpoint at localhost:8000/v1 for Aider, Codex and any OpenAI client, plus the Anthropic API at localhost:8000 for Claude Code.