Anthropic’s prompting guide for Claude Opus 5.5 advises developers to test multiple attempt levels alternatively of reusing the environment used in Claude Opus 5. It additionally suggests conversation applications eliminate system-prompt lines instructing Claude to think carefully before responding.
Opus 5.5 launched Sept. 22. The guide says Opus 5 prompts should motionless activity fine without edits, and that the company’s Opus 5 direction is a sensible location to start.
When the business tested this in a conversation product, removing a think-carefully row made replies commencement sooner, alongside “no apparent decrease in the norm of the reply.”
Opus 5.5 Defaults To Medium Effort
Opus 5.5 runs at average attempt by default, one flat below the elevated default on Opus 5.
Thinking can’t be turned off on Opus 5.5, and requests that try get an error. Opus 5 allowed that at elevated attempt or below. The guide calls attempt the archetypal environment to modify whenever balancing norm against speed and cost.
In Anthropic’s testing, Opus 5.5 at average attempt matches or equal surpasses Opus 5 at elevated attempt whenever it comes to coding and cognition work.
To cut thinking, the guide says to try lesser attempt settings before immediate changes, and preserve the higher levels akin xhigh and max for tasks anywhere they really create a difference in quality.
Anthropic Suggests Reconsidering “Think Carefully” In Chat
The guide’s logic for suggesting conversation apps reconsider think-carefully lines is that Opus 5.5 decides for itself how much to think, alongside attempt as the chief control.
Anthropic’s own docs contain that benevolent of line. For Claude Opus 4.7, whenever attempt has to remain low to keep responses quick, its page on effort suggests adding direction specified as “This project involves multistep reasoning. Think carefully before responding.”
Agent Guidance Covers Time Budgets And Pasted Text
For delegate teams, the guide recommends using a period prosperity according to how lengthy tasks power take. With Opus 5.5, the application helps by tracking elapsed time.
Anthropic’s tests have shown that small delegate groups, whenever using period signals, complete investigation tasks quicker than a sole delegate operating without signals. Even under period budgets, these groups managed to accomplish answer norm comparable to solo agents. The guide emphasizes that the period prosperity is fair a suggestion, and environment a strict timeout can be quite helpful. The example power execute slightly small thoroughly under pressure.
For users who input content from sources akin emails, the guide suggests using tags alongside a random ID and adding a scheme note on how to oversee tagged text. This method encourages careful handling and, since tags are plain content that can be copied, they provision lone one tier of safety against immediate injection.
5 Claude Opus 5.5 Prompting Tips From Anthropic
What Anthropic says to retest, remove, and add whenever prompting Opus 5.5
-
Retest attempt settings
Don’t assume your Claude Opus 5 environment is motionless the correct one for Opus 5.5.
-
Lower attempt before prompting for small thinking
Anthropic says attempt is the archetypal environment to modify whenever balancing quality, speed, and cost.
-
Consider removing ‘think carefully’ in conversation prompts
In Anthropic’s chat-product testing, removing that education made replies commencement sooner.
Anthropic stated there was ‘no apparent decline’ in answer quality.
-
Use period budgets for predictable tasks
For multiagent workflows alongside a sensible project estimate, Anthropic recommends passing elapsed period against that budget.
The prosperity is advisory, not a difficult stop.
-
Tag pasted content from exterior sources
Anthropic recommends marking pasted satisfied and telling Claude how to treat it.
It describes this as one guardrail against immediate injection, not a complete defense.
Anthropic says existing Opus 5 prompts should motionless activity well. These are Opus 5.5-specific tuning suggestions.
Why This Matters
Several sections of the Opus 5.5 guide ask developers to re-check setup carried complete from older models, specified as workarounds for study charts and screenshots, or instructions written for operating alongside thinking off. The Fable 5.1 guide, which Roger Montti covered before this month, asked developers to revisit formatting rules.
An app that never sets attempt runs at average on Opus 5.5 without any code change.
Looking Ahead
Using an output cap (max_tokens) for Opus 5 alongside ‘thinking off’ power cut responses off on Opus 5.5. The guide mentions that ‘thinking’ uses several of that cap equal whenever you don’t see it.
For frontend projects, the Opus 5.5 guide recommends environment apparent styles to evade defaults akin lotion backgrounds and pill-shaped buttons. Vague requests to steer apparent of a generic AI appearance frequently fair toggle one default for another.
Featured Image: Samuel Boivin/Shutterstock
