PRIVATE SYSTEM / 01
DEEN
CONTACT
All insights
Automation

Claude Sonnet 5: Medium vs. Low Effort — Which Should Your Automation Use?

July 30, 2026

A practical look at Claude Sonnet 5's effort parameter: what medium and low actually change, and which one fits each step of your n8n automation.

When you build an automation with Claude — whether that's an n8n workflow qualifying leads, drafting follow-up emails, or triaging support requests — one setting quietly decides how much you pay and how fast you get an answer: the effort parameter. Claude Sonnet 5 supports five levels, but for most day-to-day automations the real decision comes down to two: medium and low.

Low effort is built for speed and volume. It uses the fewest tokens, responds fastest, and costs the least per request. It's the right choice for simple, repetitive steps: classifying an incoming message, extracting a field from a form submission, deciding which of three canned replies to send. If your automation runs hundreds or thousands of times a day and each step is genuinely simple, low effort keeps the bill and the latency down without hurting the result.

Medium effort trades a bit of that speed for noticeably more careful reasoning. It's the setting for steps where the model actually has to think — summarizing a longer conversation before handing it to a human, drafting a personalized reply instead of a template, or deciding between several possible next actions in a workflow. According to Anthropic's own guidance, Sonnet 5 at medium effort performs comparably to the previous generation (Sonnet 4.6) at its highest setting — so it's a genuine cost-saving step down, not a quality compromise, for most agentic and reasoning-heavy tasks.

How to choose in practice: look at each step in your automation on its own, not the workflow as a whole. A single n8n workflow can — and often should — mix both: low effort for the classification node that decides whether something is a support ticket or a sales lead, medium effort for the node that then drafts the actual reply. Treating effort as a per-step decision, rather than a single global setting, is usually where the real cost savings show up.

A simple rule of thumb: if a human doing the same step wouldn't need to think twice about it, use low. If it's a step where you'd want a careful, well-reasoned answer, use medium.

This distinction matters more than it might seem for anyone running Claude inside a business process — it is very literally the difference between a system that is unnecessarily slow and expensive, and one that spends its budget only where the extra reasoning actually earns its keep. Getting the effort level right on each step of an automation is a small, easy improvement that pays for itself immediately, and it's exactly the kind of design decision we build into every automation system we deliver.

AIn8nClaude