Who makes the models
In one line: there are about five names worth knowing, they leapfrog each other every few months, and the useful thing to learn is what KIND of choice you are making — not which one is ahead today.
Why this lesson does not list the best model
Because that list would be wrong within weeks. Model names, prices, speeds and scores change constantly, and a page that told you "X is best at code" is actively harmful once it stops being true.
So: the names, what each house is generally known for, and a link to their own page — which is the only version that is current by definition.
The main ones
Anthropic — Claude. Known for long, careful reasoning and for being pleasant to work with on writing and code. Claude Code, which AIOS can drive directly, is theirs. → Models and pricing
OpenAI — GPT / ChatGPT. The most widely used, the biggest ecosystem of tools and plugins built around it, strong all-rounder. → Models
Google — Gemini. Very large context (you can hand it a great deal at once), tightly tied into Google's own products, and a generous free tier. → Models
Meta — Llama. The big one you can download and run yourself. Free to use, including commercially, within its licence. → Llama
Mistral — Mistral / Mixtral. European, strong small models, several fully open. Good when you want something capable that runs on modest hardware. → Models
Also worth knowing the names of: xAI (Grok), DeepSeek, Qwen (Alibaba), and Microsoft Copilot — which is mostly OpenAI's models wearing Microsoft's clothes.
The distinction that actually matters
Not "which is best". This one:
Closed models run on somebody else's computers. You send your text to them. They are usually the most capable, they cost money per use, and they can change or disappear under you.
Open-weight models you can download. They run on your own machine, your text never leaves it, they are free to run, and they are generally a step behind the best closed ones. "Open weights" is not quite the same as open source — you get the model, not always the recipe.
That single choice — someone else's computer, or mine — decides more about how you work than any benchmark. There is a whole lesson on it later.
Sizes, in plain terms
Every house sells roughly three tiers, whatever they call them:
- Small / fast / mini — cheap, quick, fine for tidying, sorting, summarising, extracting. Most of what you do.
- Mid — the sensible default.
- Large / reasoning / thinking — slow and expensive, noticeably better at hard problems, genuine analysis and tricky code.
Using the biggest for everything is the commonest and most expensive mistake. The next lesson is about choosing.
Try this
Open two of the links above and find the current model list. Notice how recently they changed. That is the reason this lesson links out instead of listing — and it is a habit worth keeping: check the vendor's page, not an article about it.
---
Next: Choosing a model for the job · All lessons