lfm2.5-1b-4bit
Run LFM2.5 1.2B on a Mac
Liquid's hybrid conv+attention architecture keeps the KV cache tiny, so long chats don't balloon memory the way same-size transformers do. Under 1 GB resident on the M4 Pro run, so it runs on any Apple Silicon Mac.
1.2 B hybrid conv+attention · 4-bit8 GB Macs · long chats on small memory · quick Q&A
Liquid AI 125K context Base model intelligence 2.3 AA
One command
rapid-mlx serve lfm2.5-1b-4bit
Weights (0.66 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="lfm2.5-1b-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 | 314.0 tok/s | 0.8 s | 1.8 GB | — | 0.66 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: 618.2 tok/s aggregate.
Community runs, one page per Mac: Apple M1 Max.
Will it fit your Mac?
Measured 1.8 GB Metal peak (0.9 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
lfm2.5-2.6b-4bit— LFM2.5 2.6B — same architecture, noticeably smarter
FAQ
How much memory does LFM2.5 1.2B need on a Mac?
Measured 1.8 GB Metal peak (0.9 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 LFM2.5 1.2B on Apple Silicon?
We measured 314.0 tok/s on a Mac mini M4 Pro · 48 GB, rapid-mlx 0.15.4, 2026-10-02; peak memory 1.8 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 LFM2.5 1.2B locally?
rapid-mlx serve lfm2.5-1b-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.