Article
Best Practices for Using Claude Opus 4.7 With Claude Code
claude.com
- To get the most out of Opus 4.7 in Claude Code, we’ve found it’s helpful to treat Claude more like a capable engineer you’re delegating to than a pair programmer you’re guiding line by line:
- Specify the task up front, in the first turn. Well-specified task descriptions that incorporate intent, constraints, acceptance criteria, and relevant file locations give Opus 4.7 the context it needs to deliver stronger outputs. Ambiguous prompts conveyed progressively across many turns tend to reduce both token efficiency and, sometimes, overall quality.
- Reduce the number of required user interactions. Every user turn adds reasoning overhead. Batch your questions and give the model the context it needs to keep moving.
- Use auto mode when appropriate. For tasks where you trust the model to execute safely without frequent check-ins, auto mode cuts cycle time. It’s an especially good fit for long-running tasks where you’ve provided full context up front. Auto mode is now available in research preview for Claude Code Max users—you can toggle it on using Shift+Tab
xhigh(default, recommended): The best setting for most coding and agentic uses. It has strong autonomy and intelligence without the runaway token usage that max can produce on long agentic runs.max: Squeezes out additional performance on genuinely hard problems, but shows diminishing returns and is more prone to overthinking. Use it deliberately for tasks like testing the model’s maximum ceiling in evals and for extremely intelligence-sensitive and non-cost-sensitive uses.- We’ve set the default effort level for Opus 4.7 to xhigh because we believe it’s the best setting for most coding tasks. If you’re an existing Claude Code user but you haven’t manually set your effort level, you’ll be upgraded to
xhighautomatically. You can still adjust your effort manually. - Adaptive thinking has improved meaningfully in this release—in particular, Opus 4.7 is less prone to overthinking. If you want more control over the thinking rate, prompt for it directly: • If you want more thinking, try something like, “Think carefully and step-by-step before responding; this problem is harder than it looks.” • If you want less thinking, try something like, “Prioritize responding quickly rather than thinking deeply. When in doubt, respond directly.” You’ll save tokens but may lose some accuracy on harder steps.
- Response length is calibrated to task complexity. Opus 4.7 isn’t as default-verbose as Opus 4.6. You can expect shorter answers on simple lookups and longer ones on open-ended analysis. If your use case relies on a specific length or style, state that explicitly in your prompt. We find that positive examples of the voice you want work better than negative “Don’t do this” instructions.
- The model calls tools less often and reasons more. This produces better results in many cases. If you want more tool use (say, more aggressive search or file reading during agentic work), provide guidance that explicitly describes when and why the tool should be used.
- It spawns fewer subagents by default. Opus 4.7 tends to be more judicious about when to delegate work to subagents. If your use case benefits from parallel subagents (for example, fanning out across files or independent items), we recommend spelling that out. For example: Do not spawn a subagent for work you can complete directly in a single response (e.g., refactoring a function you can already see). Spawn multiple subagents in the same turn when fanning out across items or reading multiple files.
- Opus 4.7 performs better on long-running tasks than prior models. This makes it a good fit for tasks where supervision used to be the bottleneck, like complex multi-file changes, ambiguous debugging, code review across a service, and multi-step agentic work.