Prompting Claude Opus 5.5
Anthropic's guide covers the prompting patterns specific to Claude Opus 5.5, focusing on behavioral differences from Claude Opus 5 and the harness patterns that address them: effort calibration, thinking behavior in API integrations and chat, progress updates, unattended and multiagent tasks, safeguard refusals, frontend design, complex visual inputs, multi-app workflows, and pasted text in user messages. It is a companion to separate docs on the model's capabilities and API changes and to Anthropic's general prompting best practices.
On migration and performance, Claude Opus 5.5 generates output tokens more than 30 percent faster than Claude Opus 5 and tends to finish the same task with fewer tokens. Existing Claude Opus 5 prompts should perform well without changes, and the patterns from the earlier "Prompting Claude Opus 5" guide remain a reasonable starting point.
The guide is organized as a symptom-to-section map. If you're unsure which effort level to run, or turns run longer and cost more than they did on Opus 5, see "Calibrate effort." If an Opus 5 integration ran with thinking disabled, see "Prompts written for thinking disabled." If an unattended agent stops partway through a long task after reporting progress, see "Unattended agentic runs." If requests return stop_reason: "refusal," see "Safeguard refusals." If long agentic turns look silent or you want updates at predictable points, see "User-facing progress updates." If an agent working across several connected apps misses information the task didn't point to, see "Explore context in multi-app workflows." If you run a team of agents and want it to finish sooner, see "Time signals for multiagent harnesses." If replies in a chat application start slowly because the model thinks at length first, see "Thinking instructions in chat system prompts." If the model follows instructions that arrived inside text a user pasted, see "Mark pasted text in user messages." If answers about dense charts, diagrams, or screenshots miss detail, see "Tools for complex visual inputs." And if frontend output looks generic, see "Frontend design defaults."
On capabilities relevant to prompting: the model is strongest at agentic coding and code review, particularly multistep work in a real repository such as carrying a change through a large code base until its tests pass. In Anthropic's testing, at its default medium effort, the model matched or beat Claude Opus 5 at high effort on such tasks, in fewer steps and with fewer tokens. It also sustains long-running autonomous work better than Claude Opus 5 — for example, multi-hour audits and migrations of large code bases run end to end with parallel subagents and little oversight. Early testers reported stronger code review as well, with more bugs caught than on Claude Opus 5, fewer false alarms, and plain-language explanations of its changes.
On knowledge work, the model is much less likely to state an incorrect figure or cite the wrong source, and it is better at financial modeling tasks such as building a financial model and one-page summary.