Lagoon · Model Sommelier
The Model Sommelier.
Tell it how much memory your machine has. It tastes the whole frontier — every open model worth your time — and pairs you with the best one you can actually run. No guesswork, no buying the wrong thing.
How much memory does your machine have?
Do not know? On a Mac: Apple menu → About This Mac. On Windows: Settings → System → About. We assume about 75% of memory is free for models; the rest runs your system. Under 8GB, nothing on the frontier runs comfortably yet.
Choose your memory to find your pairing 🍾
Your pairing
Qwen3-14B ★
The sweet spot: genuinely smart, and it fits in 9GB.
Needs ~9GB · Intelligence 37/70 · Tier Base · Ships with Lagoon Micro
Best machine for it: Mac mini M6 · 16GB (~$899) — see all machines
Join the waitlist — Lagoon Micro runs thisSee machines that run itThe frontier, in full
Every model on the chart, as plain text — the same data the chart draws, for readers and machines that do not run JavaScript. Ordered by intelligence, smartest first.
| Model | Memory needed | Intelligence / 70 | Ships with Micro |
|---|---|---|---|
| DeepSeek R1 671B | ~380 GB | 62 | — |
| Qwen3-235B-A22B | ~130 GB | 58 | — |
| gpt-oss-120B | ~65 GB | 54 | — |
| Qwen3.8-27B | ~18 GB | 48 | — |
| Gemma 4 31B | ~19 GB | 46 | — |
| Qwen3-14B | ~9 GB | 37 | ★ |
| Phi-4 14B | ~9 GB | 34 | ★ |
| gpt-oss-20B | ~12 GB | 34 | — |
| Gemma 4 12B | ~8 GB | 32 | — |
| Granite 4.1 8B | ~5 GB | 26 | ★ |
Intelligence scores are approximate blends of public benchmarks (Oct 2026) — the same estimates as the Machines page. Memory is the room a model needs to run well. A bigger machine always runs smaller models too: the pairing is the best one that fits, not the only one.