Claude Opus 5 closed last year's SDK gaps — not this year's

Claude Opus 5 improved on the SDKProof benchmark and now handles newer API changes in Vercel AI SDK 7 and Zod 4.

It still struggles with the latest breaking changes in Prisma 7 and Next.js 16, showing the newest releases remain hard for coding models.

The benchmark uses compiler-checked tasks against real packages, and a bug affecting one Prisma task was fixed.

The results suggest models quickly learn older SDK revisions, but performance lags on the newest major library updates.