I'm seeing a lot of conversations, analysis and opinions/conspiratory theories around the interwebs so I decided to put a summary of everything. It all indicates there were defaults changes to ClaudeCode that affected people's perception of "4.6 is nerfed", not that the actual model or the quality of 4.6 itself was nerfed. If that changes anything about the conspiracy theorists around or not, make your own conclusion (at least now with the data/references)
The article splits three things: default behavior (adaptive thinking, effort), UI redaction of reasoning, and whether external benchmark charts are comparable. For people who live in Claude Code daily: did you notice the read-before-edit shift or more “edit without context” before 4.7, and was visible thinking part of how you debugged failures?
Great Q. Speaking from my own experience: I use claude code daily and ALWAYS set to max effort (I use 20x). I've noticed improvements on 4.7 from 4.6, marginally, but no regression. That's my experience froms setting the effort always to what I want manually for every session.
reply