Lesson 2 of 7

The open-weights families

6 min read

If you're going to run an open model, which one? A few big families lead the pack — so which fits your task?

A handful of families

Open weights isn't one model — it's several families, each from a different maker, each in a range of sizes. The names you'll meet most: Llama (Meta), DeepSeek, Mistral, Qwen (Alibaba), and Gemma (Google). They overlap a lot, but each leans toward something — general chat, coding and math, running lean, many languages, or staying tiny.

A few open families — Llama, DeepSeek, Mistral, Qwen, Gemma — cover most of what you'd want to run.

Pick by the job, not the logo

You don't need to memorise them. Each family ships in several sizes, from tiny ones for a laptop to large ones for a workstation, and new releases arrive often. When you're choosing, start from your task — coding, chatting, a language you use — and from what your machine can handle. The next lessons show how a model's size decides whether it even fits.

Choose a family by your task and your machine — sizes and releases change, the fit is what matters.

Don't chase the leaderboard. For most everyday use, any recent mid-size model from these families is more than good enough — the 'best' one is the one that runs well on your hardware.