Heavy Claude Code user here (Max20, running 10-15 subagents in parallel most days). The "nerfed" feeling is real but I think it's mostly rate limiting rather than actual model changes. When you push the API hard enough, you get throttled to a lower priority and responses feel slower and lazier. Switch to a fresh session or wait an hour and suddenly it's "smart" again. That said, actual model version bumps do change behavior significantly. Opus 4.6 handles tool calls differently than 4.5 did and it broke a couple of my automation workflows. That felt like nerfing but was just a new model. Account type 100% matters. I've been seriously considering a second Max20 sub because running 5 agents on two accounts is more stable than 10 on one.

does everyone always have access to the same model with same capabilities, or could it be nerfed for certain users only, depending on usage patterns or account type?