Anthropic has released a dedicated guide covering prompting patterns specific to Claude Opus 5.5. The model generates output tokens more than 30 percent faster than Claude Opus 5 and typically completes the same tasks using fewer tokens.
Model ReleasesAnthropicClaude Opus 5.5
Anthropic Releases Prompting Guide for Claude Opus 5.5
A key feature of the new model is the "Effort" setting, which serves as the primary control for the model's thinking process. Since thinking is always active in Opus 5.5, users can adjust this setting to trade off intelligence, latency, and cost. Anthropic's testing indicates that the default "medium" effort level for Opus 5.5 matches or exceeds the "high" effort level of Claude Opus 5 on coding and knowledge-work evaluations.
For developers building agentic workflows, the guide suggests several optimization strategies:
- Thinking Display: Using
display: "updates"allows clients to receive short summaries of intermediate notes, preventing the appearance of silence during long reasoning turns. - Iterative Updates: Adding a system prompt instruction to "treat earlier answers as done" can reduce redundant thinking on follow-up turns.
- Resource Budgeting: Providing the model with a time budget (e.g., "elapsed 340s / 1200s") helps the model pace its work to finish within specified constraints.
Additionally, Claude Opus 5.5 demonstrates improved resistance to indirect prompt injection via tool results and web content. To further enhance security when users paste external content, the guide recommends wrapping pasted text in tags containing a unique, random ID.
Sources
- Prompting Claude Opus 5.5 （Hacker News Frontpage, 2026-09-28）