Transcription models that run on 24 GB

39 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.

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 turn recorded speech into text. They are metered by the minute or the second of audio, so cost follows the length of the recording and not the difficulty of it. What separates them is languages covered, whether they mark who is speaking, and whether they run in real time or only on a finished file — a model that is excellent on a podcast may be unusable on a live call.

Wider

Transcription models Models that run on 24 GB