How to Use Claude Sonnet 5.5 for Every Task: A Prompt Guide by Effort Level
Claude Sonnet 5.5 introduced something most AI models don't expose: a dial. You can tell it exactly how hard to think. At low effort, it answers in seconds and costs almost nothing. At max effort, it reasons through problems at a level that matches or beats Opus 5.5 on many benchmarks — at a fraction of the cost.
The catch: most people don't know which level to use when. This guide fixes that.
---
The Four Effort Levels
| Level | What it does | When to use it | Speed | |---|---|---|---| | Low | Direct answer, no deliberation | Simple lookups, formatting, drafting | Fastest | | Medium | Standard reasoning, most tasks | General writing, summaries, moderate analysis | Fast | | High | Extended thinking, checks work | Complex analysis, multi-step logic, code review | Slower | | Max | Full reasoning chains, adversarial self-checking | Hard problems, research synthesis, architecture decisions | Slowest |
The default in most interfaces is medium. You switch levels by explicitly stating the effort in your prompt.
---
How to Set Effort Level in Your Prompt
The cleanest approach is a direct instruction at the start or end of your prompt:
In the Claude API, you can also set thinking: {type: "enabled", budget_tokens: N} where N controls the token budget for reasoning (higher N = more thorough). In the Claude.ai interface, a "More Thinking" option in settings enables extended thinking for the session.
---
Low Effort — Fast, Simple, Efficient
Use low effort for anything where the right answer is clear and speed matters. Formatting, summarizing, light rewrites, fact lookups, translation.
Copy-paste prompts:
"Low effort: Convert this list of features into a two-column comparison table: [paste your list]"
"Low effort: Rewrite this paragraph in a more formal tone, keep every fact, change only the voice: [paste paragraph]"
"Low effort: Give me 10 subject line options for a cold outreach email about enterprise cloud storage. Vary the angle (curiosity, pain, social proof)."
"Low effort: Translate this UI error message into clear non-technical language a first-time user can act on: [paste message]"
"Low effort: Extract all the action items from this meeting transcript and format them as a bullet list with owner and deadline columns: [paste transcript]"
---
Medium Effort — The Default Mode
Most tasks belong here. Medium effort handles general writing, standard code tasks, summaries, and moderate analysis without burning extra time or budget.
Copy-paste prompts:
"Write a 400-word product description for a standing desk for home offices. Lead with the core benefit, use specific numbers (weight capacity, height range), and end with a direct call to action. No jargon."
"Review this Python function for correctness and suggest one refactor to improve readability. Don't rewrite the whole function — just show the relevant section:
python
[paste function]
"Summarize this 2,000-word article in 5 bullet points, each under 20 words, written for a CFO who has no domain knowledge: [paste article]"
"Write three variations of a LinkedIn post about [topic]. Variation 1: data-led opener. Variation 2: personal story hook. Variation 3: contrarian take. Each under 150 words."
"Draft a three-paragraph response to this customer complaint. Acknowledge the issue, explain what happened, and offer a specific remedy — no platitudes: [paste complaint]"
---
High Effort — Complex Analysis and Multi-Step Logic
High effort is for problems where the first answer is probably not the best answer. Code architecture, multi-factor analysis, writing that needs structural coherence across sections, argument evaluation.
Copy-paste prompts:
"High effort: Analyze this business plan for logical gaps. I want you to steelman the plan first, then list every assumption that could be wrong and rank them by the damage each failure would cause: [paste plan]"
"High effort: This SQL query is returning wrong row counts on joined tables. Trace the logic step by step, identify where the overcounting is happening, and propose the minimal fix:
sql
[paste query]
"High effort: I need to choose between three database architectures for a multi-tenant SaaS product. Walk through the trade-offs for each on these dimensions: isolation, cost, query complexity, migration risk. My usage pattern is: [describe usage]."
"High effort: Review this 1,500-word essay for argument structure. Flag any claims that lack supporting evidence, identify where the logic jumps, and suggest the two structural changes that would do the most to strengthen the argument: [paste essay]"
"High effort: Audit this React component for performance issues. I care specifically about unnecessary re-renders, missing memoization, and any async patterns that could cause race conditions:
tsx
[paste component]
---
Max Effort — The Hard Problems
Max effort enables full extended reasoning chains where Sonnet 5.5 works through the problem from multiple angles, challenges its own assumptions, and synthesizes before answering. Use it for genuinely hard, consequential decisions.
Copy-paste prompts:
"Max effort: I'm deciding between two technical approaches to migrating our authentication system off a legacy JWT implementation. Approach A is [describe]. Approach B is [describe]. Walk through the risks, failure modes, rollback complexity, and team skill requirements for each. Give me your actual recommendation with the reasoning visible."
"Max effort: This code review has three changes that interact. Tell me if applying all three is safe in the same deploy, or if there's a sequencing dependency I'm missing:
[changes]"
"Max effort: I'm writing a research synthesis on [topic]. I have these five sources that partially contradict each other. Help me reconcile the conflicting findings, identify which discrepancies are methodological vs. substantive, and produce a coherent summary a specialist could cite: [paste sources/abstracts]"
"Max effort: Critique this product strategy. I want a rigorous adversarial review — assume I'm wrong about my key assumptions. What's the most damaging version of the argument against this strategy, and how should I respond to it?:
[paste strategy]"
"Max effort: I have a logic bug in this function that only appears with nested objects more than 3 levels deep. I've already checked X, Y, Z. Trace every execution path relevant to the bug and show me exactly where the state diverges from expected: [paste code + test case]"
---
Quick Decision Guide
| Task type | Effort level | |---|---| | Format, translate, rewrite | Low | | General writing, light code | Medium | | Complex analysis, architecture | High | | Hard bugs, research synthesis, strategy critique | Max | | You're not sure | Medium — upgrade if the answer feels incomplete |
---
One Rule That Changes Everything
Match effort to consequence, not complexity. A complex-sounding task that doesn't matter much (rewriting a Twitter bio) belongs at low. A seemingly simple task with high stakes (choosing a dependency for a production system) belongs at high or max. The variable is the cost of being wrong, not the difficulty of the words.
---