Skip to content
Zenteck
Latest
Prompts

Still Using 'Think Hard'? Why Your Claude Prompts Are Failing

ByAlex Silva2 min read0 comments
outdated-prompting-habits-slow-claude-opus-5
Photo: ZenteckOutdated prompting habits slow Claude Opus 5.5

Listen to article0:00 / 0:30

If you are still typing "think hard" or "think carefully" into your prompts, you are actively slowing down Anthropic's newest flagship model. According to a guidance report published by The Decoder, the release of Claude Opus 5.5 breaks several legacy habits carried over from older reasoning architectures.

Older models required explicit coaching to trigger deep analysis. Opus 5.5, however, dynamically scales its reasoning depth based on the task and your selected effort settings. Internal tests by Anthropic show that commands featuring these filler phrases do not improve output quality, but they do delay the initial response.

How to adjust your workflow for Opus 5.5

To get the most out of the default model on Pro and Team plans, you need to change how you structure your daily instructions. Start by dropping micro-management and shifting your baseline settings.

  • Start at Medium effort: Anthropic's tests reveal that Opus 5.5 running on the default "Medium" setting matches the performance of the older Opus 5 set to "High". Save the "High" setting for complex edge cases to avoid burning extra tokens.
  • Stop asking for drafts: Instead of requesting an initial outline and asking for revisions in steps, ask for the finished document, spreadsheet, or code file directly in your first prompt.
  • Attach raw files directly: Do not waste time manually transcribing data tables or charts into text. The model's updated image reading extracts dense data accurately even at lower effort levels.
  • Define concrete styles: When generating design assets, ban specific unwanted elements (like pill-shaped buttons) rather than relying on vague terms like "generic design".

Vale a pena mudar seus hábitos de prompt?

Relearning how you write prompts takes conscious effort on Monday morning, but the payoff is immediate. Teams that previously relied on cheaper models like Sonnet at maximum effort often spent comparable amounts without gaining the reasoning depth needed to finish multi-step jobs in a single pass. By letting Opus 5.5 handle the back-and-forth execution from a single comprehensive instruction, you cut down the chat friction and get straight to the final output.

Sources

  1. Anthropic says outdated prompting habits are slowing down Claude Opus 5.5 — the-decoder.com

Frequently asked questions

Why does 'think hard' slow down Claude Opus 5.5?
Opus 5.5 automatically scales its reasoning depth for every task. Adding legacy phrases does not improve quality, but it does delay the initial response.
What is the recommended default effort level for Opus 5.5?
Anthropic recommends starting at the 'Medium' effort level, which matches the performance of older models set to 'High' while saving compute costs.

AI Briefing in 3 Minutes

Get what actually matters in AI & tech delivered free to your inbox. Zero spam, 1-click unsubscribe.

By subscribing, you agree to our Privacy Policy.