Claude Code users say UX is deteriorating after Opus 4.7 — extended thinking removed

What shifted

It has been reported that users of Anthropic’s Claude Code have seen a steady decline in developer-facing quality over the past few months, and that Opus 4.7 made the trend impossible to ignore. In the release notes Anthropic removed “extended thinking” budgets — setting thinking: {"type":"enabled","budget_tokens":N} now returns a 400 error — and said adaptive thinking “reliably outperforms” the old mode in their internal evaluations. Meanwhile, earlier changes quietly rolled out in March and April: a plan-mode option to clear context was toggled off (then later restored), cache TTLs briefly dropped from one hour to five minutes, and it has been reported that third‑party apps were blocked from charging user subscription token balances without an announcement.

Why people are upset

Developers say these aren’t cosmetic tweaks. They reportedly compound into real regressions for multi-step or large-context engineering tasks — slower performance, less predictable behavior, and fewer levers to fix things when they go wrong. GitHub issues and forum threads filled up fast, and many complaints were marked “closed” or “not planned,” which only added to the frustration. Is this what growth looks like? For some, it feels like the classic enshittification curve — growth first, then product tricks to control costs as scale bites back.

Anthropic’s position — and the human cost

Anthropic frames these moves as capacity and cost-mitigation measures during rapid adoption: it has been reported that the company is prioritizing reliability and efficiency while usage spikes. That may be true. But for paying API customers and engineers trying to ship complex systems, the net effect is a diminished quality of life — fewer options, less control, more surprises. The emotional moment here is plain: people feel pushed out of a tool they relied on, and that sting is louder than any polite engineering rationale.

What comes next

There’s no clear fix beyond Anthropic deciding otherwise. Some changes were reversed after pushback; others remain. The community is watching closely, and it has been reported that conversations on social media and in issue trackers are keeping pressure on the company. Will Anthropic prioritize developers’ control again — or will adaptive defaults and cost playbooks win out? Time will tell, and engineers are already updating their backups.

Sources: blog.matthewbrunelle.com, Lobsters