lfm2.5-2.6b-4bit
Run LFM2.5 2.6B on a Mac
Our pick for 8 GB Macs: a 1.6 GB download, and Liquid's hybrid conv+attention architecture keeps the KV cache small, so context growth is cheap.
2.6 B hybrid conv+attention · 4-bit8 GB Macs — the default pick · everyday chat
Liquid AI 125K context Base model intelligence 11 AA
One command
rapid-mlx serve lfm2.5-2.6b-4bit
Weights (1.6 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="lfm2.5-2.6b-4bit". Claude Code:
ANTHROPIC_BASE_URL=http://localhost:8000 (no /v1 suffix) —
see the Claude Code guide.
Current-version run status
Attempted on · Mac mini M4 Pro · 48 GB
Run record
| Machine | Decode | First token | Peak memory | Cold boot | Weights |
|---|---|---|---|---|---|
| Mac mini M4 Pro · 48 GB rapid-mlx 0.15.5 · 2026-10-03 · not measured raw data + run record | — | — | — | — | 1.6 GB |
Each row cites its own run; numbers are never copied across machines. Mac mini M4 Pro · 48 GB: rapid-mlx 0.15.5, 2026-10-03 — raw data + method.
Not yet measured on rapid-mlx 0.16.0. LFM2.5 2.6B has not been measured on rapid-mlx 0.16.0 yet. Our 2026-10-03 run (rapid-mlx 0.15.5, 48 GB M4 Pro) produced no numbers; its record is published with the raw data. Earlier numbers came from a pre-0.14 engine and have been removed rather than left standing. Raw data and run record →
Will it fit your Mac?
Not yet measured on rapid-mlx 0.16.0 — the weights are 1.6 GB on disk, and the KV cache grows with context on top of that.
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
lfm2.5-1b-4bit— LFM2.5 1.2B — half the footprint
FAQ
How much memory does LFM2.5 2.6B need on a Mac?
Not yet measured on rapid-mlx 0.16.0 — the weights are 1.6 GB on disk, and the KV cache grows with context on top of that.
How fast is LFM2.5 2.6B on Apple Silicon?
LFM2.5 2.6B has not been measured on rapid-mlx 0.16.0 yet. Our 2026-10-03 run (rapid-mlx 0.15.5, 48 GB M4 Pro) produced no numbers; its record is published with the raw data. Earlier numbers came from a pre-0.14 engine and have been removed rather than left standing.
How do I run LFM2.5 2.6B locally?
rapid-mlx serve lfm2.5-2.6b-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.