Images models that run on 24 GB

112 models with published weights that fit in 24 GB — a 24 GB card, or a Mac with 24. Counted at four-bit quantisation, weights only, leaving the machine about a third of its memory and a gigabyte for context: room for roughly 24 billion parameters. A mixture of experts is counted in full, because it is held in full even though only a few experts compute. 112 in all, a hundred to a page; this is page 1 of 2.

A 24 GB graphics card or a Mac configured with 24. This is the first size where the well-regarded mid-weight models — the twenty-something billion parameter class — run comfortably, and where a local model starts to be a real alternative to an API for daily work rather than a demonstration.

Models that make pictures from a description, and increasingly edit the ones you give them. Priced per image, sometimes by resolution, so the rate card reads nothing like a language model's. Beyond the look of the output, the practical differences are whether it can render legible text, whether it will edit rather than regenerate, and what the licence says about selling what comes out.

Wider

Images models Models that run on 24 GB