The Column Orders.
Portico doesn't lock you into one model. Point it at any open-weight model, and it handles the rest. To make it easy to choose, we've categorized our recommended models into architectural orders based on how much RAM your machine has.
Don't see your model on this list? No problem. Portico supports any standard GGUF model file. Just drag it into the app, and it will run.
How much memory do you have?
That single number decides more than anything else about which models you can run. Find your row, then read the order below it.
Around 3B. Quick drafts, summaries, older laptops.
See the models
Around 8B. General reasoning and code — the sweet spot for most machines.
See the models
14B to 32B. Desktops, longer context, harder problems.
See the models
Around 70B. Near-frontier quality, entirely offline.
See the modelsNamed after the classical orders, in ascending order. A model larger than your memory will still load — it just reads from disk on every word, which is far too slow to use. Portico checks your machine and marks what fits.
Doric — about 3B, 8 GB RAMQuick drafts, older laptops.
Runs fast on your card (~50 tok/s). The best choice for quick questions, summarizing text, or running on machines that weren't built for heavy lifting.
Fastest here; quick questions on any machine.
The balanced default for everyday tasks.
Google's lightweight model; surprisingly sharp.
Reasoning above its weight class.
Ultra-light and fast for basic text generation.
Sees and understands images.
Ionic — about 8B, 16 GB RAMYour machine's sweet spot.
The best balance of quality and speed for general reasoning and code. If you have a modern laptop with 16 GB of RAM, this is where you should live.
Clear step up from 3B class models.
All-rounder with massive 128K context.
The previous generation; still excellent for coding.
Newer generation, thinks before answering.
The classic workhorse. Fast and reliable.
Google's model; the most natural writing style.
128K context, excellent for creative writing.
Shows its reasoning steps openly.
Specialized for programming and code generation.
Code completion and generation across many languages.
Maths only — a great study aid.
Excellent Spanish + 22 other languages.
Follows complex instructions and personas.
Corinthian — 14–32B, 32 GB RAMDesktops and workstations.
Slow on your laptop but highly capable. These models are designed for dedicated hardware and offer massive leaps in reasoning, coding, and prose.
Strongest reasoning you can run locally.
A serious coding assistant for complex repos.
Best quality-per-GB with a thinking mode.
Microsoft's mid-tier; excels at logic and reasoning.
Beautiful prose — needs ~19 GB RAM to run well.
Strong multilingual capabilities and deep context.
Built specifically for RAG and tool use.
Mixture-of-Experts; fast responses, high quality.
Near-flagship performance locally.
Deepest reasoning capabilities for complex logic.
Composite — 70B, 64 GB RAMThe flagship tier.
For high-end workstations with massive amounts of memory. These models offer frontier-class intelligence, but are completely out of reach on standard laptops.
Flagship-class. Out of reach on this laptop, but stunning on a workstation.
Top-tier open model. Exceptional at coding and math.
Massive MoE architecture. Incredible reasoning.
Images and speech.
These don't fit the column metaphor since they aren't chat models, but Portico supports them natively for multimedia creation and transcription.
🎨 Image Generation (measured on your GPU)
15s generation — best value, your default.
15s generation — highly photographic.
15s generation — the plain original model.
Ultra-fast (2s) generation for instant drafts.
84s generation @ 768px resolution.
197s @ 1024px — the absolute best quality.
Slow generation; stronger prompt following.
Stunning prompt adherence and text generation.
🎙️ Speech
Bundled engine. Supports ~99 languages for local transcription.
Maximum accuracy for complex audio and accents.
Ultra-fast local text-to-speech in dozens of voices.
Try it on your own machine.
Free and open source. Works with the open models you already trust.
On a Mac or Linux? Build it in three commands →