Anthropic says outdated prompting habits are slowing down Claude Opus 5.5

What changed

Anthropic made Claude Opus 5.5 the default AI model on all paid plans and published new guidance targeting outdated prompting habits. The company advises dropping the typical instruction to “think hard” and instead start prompts at a Medium effort level. This shift means rethinking how users interact with the model to avoid slowing it down.

Why builders should care

Outdated prompting techniques like pushing the model to “think hard” increase computation time and reduce efficiency. By defaulting to Medium effort, Claude Opus 5.5 operates faster while maintaining quality. Builders and developers who rely on Claude can reduce latency, lower compute costs, and improve user experience by adjusting prompt styles to match Anthropic’s new recommendations.

The practical takeaway

Operators should stop overloading prompts with calls for heavy reasoning right upfront. Instead, begin with moderate effort instructions to get quicker, more cost-effective responses. This change pressures prompt engineers to rethink workflows and refactor existing prompts that nudge the model to work harder unnecessarily. It also signals that optimizing AI efficiency happens at the prompt level, not just behind the scenes in model upgrades.

What to watch next

It will be important to monitor whether this new prompting approach affects Claude’s output quality in real-world applications. Other AI providers may follow suit by tuning model defaults or offering prompting guides focused on efficiency. Also watch if Anthropic revises their guidance further to balance speed and depth of reasoning as users adapt. Prompt engineering is shifting from a trial-and-error endeavor toward a more disciplined, efficiency-driven practice.

AI Quick Briefs Editorial Desk