The first wave of developer feedback on Claude Opus 4.7 was rough. Reports of degraded coding performance compared to Opus 4.6 circulated quickly - worse at tracking context across multiple files, more prone to hallucinating function signatures, less reliable on complex instruction chains. A lot of developers quietly switched back to Opus 4.6 and stayed there.
A few weeks in, those complaints have thinned noticeably. Developers who skipped Opus 4.7 entirely at launch are now asking whether something changed.
The ambiguity here is real. Anthropic doesn't publish changelogs for backend model adjustments - and they do make them. Post-launch tweaks to sampling parameters, system prompt handling, or serving infrastructure can meaningfully shift how a model behaves without any public announcement. It happened with GPT-4 in 2023, when OpenAI faced accusations of quietly degrading the model's reasoning after users noticed regression. Whether Opus 4.7's early problems were a deployment issue that got patched, or whether the initial criticism was partly overstated, is difficult to confirm from the outside.
What's clear is that Sonnet 4.6 has become many developers' default for coding work regardless of how Opus performs. Sonnet is faster, cheaper, and the quality gap between mid-tier and flagship Claude models has narrowed with each generation. For Opus 4.7 to justify its price premium in a coding workflow, it needs to demonstrably outperform Sonnet on the tasks that actually slow you down - complex multi-file refactors, debugging subtle logic errors, maintaining consistency across a large codebase. That's a case-by-case determination, not a universal verdict.
Developers who shelved Opus 4.7 at launch have a reasonable window now to retest it on their specific stack. Launch-day impressions of a large model aren't always its final form.