One spec decides what AI a computer can run: memory. How much you have decides which models fit. How fast it is decides how quickly they answer. Pick a tier, pick a machine — that's the whole method.
Catalog current as of November 2026 — machines, models, and prices re-checked as the market moves.
Explore intelligence
Every open-weight model worth your time, plotted: smarter as you go up, hungrier for memory as you go right. The dashed curve is the Pareto frontier — the best intelligence at each memory size. Tap a model and we'll show you the machine that runs it.
Tap a dot — the card flips to reveal your pairing. ★ = ships with Lagoon Micro. Intelligence is an approximate blend of public benchmarks (Oct 2026); memory is the room a model needs to run well.
The short version
The whole page on one table. Pick your tier by the models you want to run; the pick is the machine we'd buy in it. Tap a row to jump straight to that tier.
Value: GMKtec EVO-X3 · 128GB Fastest: Mac Studio M5 Max · 128GB
~$3,800–$5,099
Frontier
192GB+
200B+ — Qwen3-235B-A22B, DeepSeek R1
Mac Studio Ultra · up to 512GB
~$5,499
Windows laptops top out at Standard — no laptop GPU holds more than 24GB. For 70B models in a portable machine, you need unified memory (Apple Silicon, AMD Strix Halo).
The catalog
Every machine that matters, filterable like it should be. It opens on Macs, laptop-first — switch a section to Desktop for the desktop picks. The full catalog is one tap away (Maker → All).
Prices are November 2026 approximations, and the market is moving — "Check price" is the real quote. Buy links below are affiliate links; we may earn a commission at no cost to you.
Tier
Maker
Complete machines
Out of the box, ready to run. Buy it, plug it in, your AI lives there.
Base — runs Lagoon Micro
8–14B models: sharp, fast everyday AI. Where almost everyone should start.
Form
16 GB unified
Mac mini M6 · 16GB
Apple M6 · desktop
The new base mini — Apple's M6 makes it the fastest cheap local-AI box ever sold. Silent, sips power, runs Micro's models all day.
The biggest open models, the ones that rival the big cloud AIs — running on your desk, answering to you.
Form
64 GB unified
Framework Desktop · 64GB
AMD Ryzen AI Max · desktop
Modular, repairable, yours in a way sealed boxes aren't. One desktop from 32GB to 192GB — pick the memory your models need. The 192GB config is Linux-only.
Memory speed
256 GB/s
Usable for models
~48 GB
Sold
Direct from Framework
OS
Windows / Linux
Memory
⚠ 64GB won't run gpt-oss-120B (needs ~65GB) — 192GB runs every Reserve model.
Beyond Reserve. The biggest open models in existence — 200B parameters and up — on hardware that lives in your home. Some of these pay us nothing. They're here because a catalog of what's possible should be complete.
512 GB unified
Frontier
Mac Studio Ultra · 512GB
Apple M5 Ultra · desktop
Half a terabyte of memory in a box the size of a small cake. Runs the largest open models that exist, alone, in silence. The M5 Ultra brings the 512GB config back — the largest memory pool in a personal computer.
Full disclosure: this page contains affiliate links to Amazon, B&H, and manufacturers. If you buy through them, Lagoon may earn a commission at no extra cost to you. We list the full catalog, not just machines that pay us. Some links — Framework — are direct and pay nothing. As an Amazon Associate I earn from qualifying purchases. Prices shown are approximate (Nov 2026) and move fast in a chip shortage; the retailer's price is the real one. Many machines are sold in smaller memory configs that unlock less — check the memory before you buy.