Extended thinking in Claude Opus is a powerful feature for complex reasoning tasks. But it comes with a tradeoff: higher cost and slower response times. The truth is, most daily AI work doesn't need extended thinking at all. You can get the same quality results with Claude Sonnet and the right effort setting.

This isn't about settling for less. It's about matching the tool to the task. Code reviews, email drafts, report summaries, data formatting, simple refactors—these are routine tasks where Sonnet shines. Architecture decisions, complex debugging, multi-file refactors, system design—these need Opus plus extended thinking.

The Claude Model Tier System

Claude offers three model tiers: Haiku, Sonnet, and Opus. Each has strengths. Haiku is fastest and cheapest for simple tasks. Sonnet balances speed, cost, and capability. Opus delivers maximum reasoning power for complex problems.

Extended thinking is available primarily in Opus. It allows the model to think through problems step-by-step before generating a response. This is invaluable for complex reasoning, but it adds 30-60 seconds of processing time and increases costs significantly.

The best workflow isn't always reaching for the most powerful tool. It's choosing the right tool for the job at hand. Most daily AI tasks fall into the routine category where Sonnet with low effort delivers identical quality to Opus with extended thinking.

When Sonnet Outperforms Opus

Here's the counterintuitive truth: for routine tasks, Sonnet with low effort often matches or exceeds Opus with extended thinking. Why? Because extended thinking is optimized for deep reasoning, not quick pattern matching. Routine tasks rely on pattern recognition and clear instructions, not step-by-step logic chains.

The cost difference matters too. Sonnet costs significantly less per token than Opus. Extended thinking adds substantial token overhead for the thinking process itself. For a single code review, this might not matter. For hundreds of daily tasks across a team, the difference compounds quickly.

Tasks That Belong on Sonnet

Code reviews are perfect for Sonnet. You want quick, consistent feedback on common issues: naming conventions, structure, obvious bugs, style consistency. Extended thinking won't improve this. It will just slow it down. Set effort to low and get fast, reliable reviews.

Email drafts follow the same pattern. Most emails follow standard structures: greetings, context, ask, next steps. Sonnet handles these templates efficiently. Extended thinking adds no value unless you're crafting sensitive communication requiring nuanced tone analysis.

Report summaries are another Sonnet use case. Extract key points, structure findings, highlight metrics. This is information synthesis, not complex reasoning. Sonnet with low effort delivers clean summaries without the wait time.

Data formatting tasks are ideal for Sonnet. Convert JSON to CSV, restructure arrays, clean up inconsistencies. These are transformation tasks with clear input-output rules. Extended thinking doesn't help here; it just adds latency.

Simple refactoring fits Sonnet too. Rename variables, extract functions, update imports. Standard code improvements that follow well-established patterns. Sonnet handles these quickly and reliably.

Tasks That Need Opus Plus Extended Thinking

Architecture decisions require the full power of Opus with extended thinking. These involve tradeoff analysis, long-term implications, and competing constraints. The step-by-step reasoning of extended thinking helps explore different approaches before settling on a recommendation.

Complex debugging benefits from extended thinking. When you're facing intermittent bugs, performance issues, or unexpected behavior across multiple files, the ability to reason through potential causes systematically makes a difference. Opus can trace through code paths and identify non-obvious interactions.

Multi-file refactors need Opus plus extended thinking. These involve understanding relationships, identifying ripple effects, and planning the sequence of changes. The upfront thinking pays off in fewer errors and more thorough refactoring.

System design work requires extended thinking. You're balancing requirements, constraints, scalability, and maintainability. The ability to think through different architectural options before proposing solutions leads to better design decisions.

Think of it like this: extended thinking is for exploration and synthesis. Routine tasks are about recognition and execution. Match the model and effort level to what you actually need from the AI.

How to Set Effort Level

In Claude Code, you can control effort through the configuration. Low effort prioritizes speed and cost efficiency. Medium effort balances quality with performance. High effort maximizes output quality at the cost of speed and increased token usage.

For routine tasks like code reviews, email drafts, and data formatting, low effort is ideal. You get fast, consistent results without paying for reasoning capacity you don't need. The key is providing clear instructions and good context.

In Claude.ai, the process is similar. Choose your model, then adjust the effort setting based on task complexity. Quick questions and routine work get low effort. Complex problems requiring deep reasoning get high effort or extended thinking if available.

The 80/20 Rule Applied to AI Workflows

The 80/20 rule applies perfectly to AI workflows. Roughly 80% of daily AI tasks are routine: code reviews, emails, summaries, formatting, standard refactors. These don't need extended thinking. They need speed, consistency, and cost efficiency.

The remaining 20%—architecture decisions, complex debugging, multi-file refactors, system design—deserve the full power of Opus with extended thinking. These are the tasks where deep reasoning makes a material difference in outcomes.

The mistake is using extended thinking as the default. It's like using a sledgehammer to hang a picture frame. It works, but it's overkill. The smart approach is to match the tool to the task: Sonnet with low effort for routine work, Opus with extended thinking for complex problems.

This isn't just about cost. It's about workflow efficiency. Fast responses on routine tasks keep you in flow. Waiting 30-60 seconds for extended thinking on simple tasks breaks momentum. Save the deep thinking for when it actually moves the needle.

Practical Implementation

Start by categorizing your AI tasks. Which are routine? Which require deep reasoning? Set up presets or workflows in Claude Code and Claude.ai that match these categories with appropriate model and effort settings.

For code reviews, create a preset using Sonnet with low effort. For architecture discussions, use Opus with extended thinking. For email drafts, Sonnet with low effort. For complex debugging, Opus with extended thinking.

The key insight is that quality doesn't suffer when you match the tool to the task. A code review from Sonnet with low effort is just as thorough as one from Opus with extended thinking. The difference is speed and cost.

Monitor your usage patterns. Track which tasks actually benefit from extended thinking. You'll likely find it's a small subset of your work. Optimize accordingly. Use the powerful tools when they matter, use the efficient tools for everything else.

The goal isn't to avoid extended thinking. It's to use it strategically. Like any powerful tool, its value comes from applying it where it makes a difference, not using it everywhere by default.

Moving Forward

The Claude model tier system gives you options. Haiku for speed, Sonnet for balance, Opus for power. Extended thinking adds another dimension for complex reasoning. The art is matching the right combination to each task.

Next time you reach for extended thinking, pause. Ask: does this task actually require step-by-step reasoning? If yes, use it. If not, Sonnet with low effort will deliver the same quality faster and cheaper.

This isn't about limiting what you can do with AI. It's about being strategic with resources. Use the powerful tools when they matter, use the efficient tools for routine work. The 80/20 rule applies to AI workflows just like everything else.

Your workflow will be faster, cheaper, and just as effective. The key is matching the tool to the task, not defaulting to the most powerful option for everything.