AI Models

Thanks for the feedback. We've actually trialled some of the larger on-device models, and the interesting result was that the jump in correction quality turned out to be small, not enough to justify the extra memory and load they ask of your Mac. So for now we've deliberately kept Charm on the lighter models to keep it fast for everyone, including people on more modest machines.

That said, this is something we watch closely. As newer models arrive that are both more capable and efficient, we'll fold the good ones in for sure, and we'll keep evaluating as the on-device model landscape moves. If a bigger model ever clearly earns its place on quality, it'll make it into Charm.

Thanks for taking the time to suggest it.