Migrating: Opus 4.7 to Opus 5

A narrower gap than migrating from 4.6

Moving from this generation is a smaller jump than migrating from Opus 4.6, since the sampling fields and thinking-budget mechanism were already gone by this point — that hard break already happened one generation earlier. What actually changes here is more incremental: the cache minimum drops from this generation's outlier-high figure, fast mode becomes available again after being removed on this specific generation, and thinking's default behaviour shifts.

What's actually worth testing on this specific move

Because this generation's cache minimum was already unusually high, re-verify your caching setup after this move specifically — a prefix structured around this generation's demanding minimum will almost certainly clear the new, lower one, but it's worth confirming rather than assuming. If your integration disabled fast mode entirely because this generation didn't support it, that's a capability now available again worth evaluating on its own merits rather than left off by habit.

Thinking-default behaviour still needs its own check

Even though the hard-breaking fields from the older generation are already gone by this point, the destination model's thinking-on-by-default behaviour is still a real change worth verifying against your max_tokens budget — don't assume this migration is purely incremental just because the biggest compatibility break happened at an earlier boundary than this one. Re-check max_tokens sufficiency specifically, and confirm whether your integration's fast-mode setting, if any, should now be revisited given it's available again on the destination.

Related

See speed: "fast" is not supported on Opus 4.7 for the specific gap this migration closes, and the minimum-cacheable-prefix table for the full non-monotonic picture across the lineup.