Anthropic is advising Claude Opus 5.5 users to abandon several prompting habits carried over from earlier generations of models. According to the company, Opus 5.5, now the default model for Pro, Max, Ultra and Team plans, works differently from its predecessors, and outdated instructions may prevent users from getting the most out of it.
At the top of Anthropic’s recommendations is removing phrases such as “think deeply” or “think carefully” from prompts. The company says the model already reasons independently and adjusts the depth of its reasoning according to the task and the selected effort level. In Anthropic’s internal tests, responses without these phrases were generated faster, with no measurable decline in quality.
Such instructions date back to the early days of reasoning models, before adaptive reasoning became a standard feature. Around a year and a half ago, users could sometimes get better results or influence a model’s reasoning process by including these phrases.
Anthropic also recommends starting with the “Medium” effort level, which is the default setting for Opus 5.5. In the company’s internal tests, Opus 5.5 at Medium matched the performance of its predecessor, Opus 5, at the High level. Higher settings increase costs but do not always deliver better results. Anthropic therefore recommends switching to High only when Medium is insufficient. Users can change this setting through the model menu next to the send button.
The company also believes teams that previously relied on the cheaper Sonnet model and reserved Opus for complex projects should reconsider their approach. A more capable model may complete a task in fewer steps, partly offsetting its higher per-token price. Anthropic notes that users who regularly run Sonnet at the “Extra High” or “Max” effort levels are already paying almost as much as they would for Opus 5.5.
From Conversation to Goal-Oriented Instructions
Instead of giving instructions one step at a time, Anthropic recommends describing the desired outcome in a single message. The prompt should specify the goal, the expected output format and the circumstances in which the model should ask follow-up questions.
According to the company, Opus 5.5 is better than its predecessor at completing multi-step tasks. It can increasingly take over processes that users previously managed through back-and-forth conversations and can generate effective intermediate instructions from a single initial prompt.
Anthropic also recommends attaching diagrams and tables as images or files rather than describing them in text. The company says workarounds required by older models, such as manually entering values from charts, are no longer necessary.
Users who need a finished document or spreadsheet should request the final product directly instead of first asking for a draft. Anthropic says Opus 5.5 can often turn a single prompt into a completed deliverable. The model’s visual reading capabilities have also improved significantly. According to the company, even at its lowest effort level, Opus 5.5 reads values in complex charts more accurately than Opus 5 did at its highest effort level.
Be Specific About Unwanted Design Elements
For design tasks, Anthropic recommends explicitly identifying unwanted features instead of relying on vague instructions such as “avoid generic design.” For example, users should specify that they do not want cream-coloured backgrounds or pill-shaped buttons.
According to the company, broad instructions may simply replace one default design style with another rather than produce the intended result.
Anthropic also says that connecting Claude to users’ files and systems through integrations can lead to significantly more personalised results. This allows the model to retrieve the information it needs independently and tailor its responses to the user’s existing work environment.
