StartupCities








Claude 4.7 is like the AI equivalent of "New Coke." Something in its core recipe changed and broke the experience.
I've always used the latest model for every provider. I often felt that the new models weren't as bad as people thought (change is just hard!). But after many hours with 4.7, it's just not good.
I've tried all the usual tricks: CLAUDE.md bullying, no CLAUDE.md, max effort, low effort, small contexts, big contexts, PLAN mode, no PLAN mode, uninstalling skills and MCPs... nothing works.
4.7 writes a wordy book report for every answer. Nothing I do makes the model concise.
Its plan documents are long and confusing and yet also miss critical details.
I rewrote large parts of 4.7 code yesterday with 4.6 and it was a breath of fresh air.
If you're struggling with New Claude, I recommend downgrading to 4.6.


















