[gorbypark]

The M2 MBA can be specced up to 24GB. 32 would be awfully nice for local models, but 24 vs 16 is a decent bump in what models can be run (and at what quant). I debated a long time over the 16 vs 24 and ended up making the right choice (local AI was not in my mind at the time at all).

[turnsout]

Yes, and it's the same for the M3 MBA. If the M4 remains capped at 24, I'll still consider it, but it'll be less cut and dry. The speed boost would be nice, but literally the only time that I feel CPU-bound is when working with local models.