Prompt Engineering3 min read
Stop writing 'think hard' in Claude Opus 5.5 prompts, Anthropic says
Anthropic made Claude Opus 5.5 the default model on paid plans and published a guide showing that common prompting habits from older models slow it down. Dropping phrases like 'think hard' and starting at Medium effort gets faster results with no quality loss.

On this page
Anthropic made Claude Opus 5.5 the default model on Pro, Max, Ultra and Team plans on 14 November and published a guide warning that habits picked up from older Claude versions are holding the new model back.
What happened
Claude Opus 5.5 is now the model you get by default on every paid Claude plan. Alongside the change, Anthropic published a list of prompting tips that apply specifically to this model, because it reasons differently from its predecessors.
The biggest change is how the model handles effort. Opus 5.5 always reasons on its own and scales its depth to the task and the effort level you choose. Phrases like "think hard" or "think carefully" were useful about a year and a half ago, when users could sometimes coax more performance out of older models that way. With Opus 5.5, Anthropic says those phrases add no quality and slow the response start.
Anthropics's full list of tips covers:
- Drop "think hard" and "think carefully" from prompts
- Start at Medium effort, the default, and move to High only when Medium is not enough
- Describe the full goal, output format and when to ask questions in one message, rather than guiding the model step by step
- Attach diagrams, charts and tables as images or files instead of typing out the values
- Ask for the finished document or spreadsheet directly, not a draft first
- For design tasks, name the specific styles you do not want (for example, a cream background or pill-shaped buttons) rather than saying "avoid generic design"
- Connect Claude to your own files and systems so it can pull in what it needs
- Share your house style so the model writes in your voice
On cost, Anthropic argues that teams who regularly push Sonnet to Extra High or Max effort end up spending about as much as they would on Opus 5.5, because a more capable model can finish a task in fewer steps.
What it means for your work
- Phrases you may have added to every prompt, like "think carefully before answering", now slow things down with no benefit.
- You can stop transcribing chart values by hand. Attach the image and Opus 5.5 reads it. Anthropic says even at the lowest effort level it read dense charts more accurately than Opus 5 at the highest.
- Long back-and-forth conversations can often be replaced by one detailed opening message covering your goal, the format you want and any questions the model should ask before proceeding.
- If your team uses Sonnet at high effort for complex work, the cost difference between that and Opus 5.5 at Medium may be smaller than you expect.
What to do now
- Open any saved prompts or templates you use with Claude and remove phrases like "think hard", "think carefully" or similar.
- Check the effort level in the model menu next to the send button. Leave it at Medium unless a specific task needs more.
- For your next report or document, write one message with the goal, the format and any constraints, rather than building up through several exchanges.
- Next time you need Claude to read a table or chart, attach the file or image instead of typing out the data.
- If your team is on Sonnet at high effort for complex tasks, compare what you spend there against Opus 5.5 at Medium before your next billing cycle.
The simplest change you can make today is also the fastest: delete "think hard" from your prompts and let the model decide how much reasoning the task needs.