I'm surprised by the crazy pace at which these models have been coming out over the past few weeks. The cheaper cache here is really nice, but I think it depends entirely on your workflow. If you're constantly compacting or burning through your context, you're probably going to have a bad experience regardless of which model you're using