Anthropic’s Claude Opus 5 guide says stronger models need fewer instructions

eyishazyer · x · 2026-07-29

A reply says newer, stronger models need fewer instructions, not more.
It points to Anthropic’s official Claude Opus 5 prompting guide, where a surprising amount of advice is essentially: stop over-instructing the model.
Examples mentioned include:
don’t tell it to “double-check” if it already does that;
don’t ask it to “re-check” or “verify” unless needed, because it may overdo self-correction and waste tokens.
The practical takeaway is that prompt engineering for frontier models is shifting from heavy-handed control toward leaner, trust-the-model prompting.

More from Models