The Claude 4.8 setting nobody covered

The four-level dial that changes everything

May 31, 2026

I spent the first morning after Claude Opus 4.8 dropped doing what everyone else did. Reading the benchmark charts, nodding at the numbers, filing it under "nice, it's smarter now." Then I noticed a small thing in the interface that nobody in my feed was talking about, and it quietly reorganized how I use Claude.

It shipped with effort controls. Four levels: Low, Medium, High, Max. You set the depth before Claude starts thinking.

That reads like a throwaway UI toggle. It isn't. It's closer to the difference between asking a consultant to glance at something and asking them to actually sit down and work through it. Same person, completely different output, because you told them upfront how much it mattered.

In partnership with

Speak messy. Prompt clean.

Go on tangents. Change your mind mid-sentence. Say "um" twelve times. Wispr Flow doesn't care — it takes everything you say, strips the filler, and gives you clean, structured text ready to paste into any AI tool.

The result: prompts with the full context your AI tools need to give you useful answers. Not the abbreviated version you'd type because typing is slow.

Works inside ChatGPT, Claude, Cursor, and every app on your screen. Millions of users worldwide, including teams at OpenAI, Vercel, and Clay.

The old way versus the new way

Before 4.8, Claude applied roughly the same cognitive depth to everything. Proofreading a sentence got the same treatment as writing a pricing strategy. I had no real say in how hard it thought.

In practice that created a strange situation. I'd get over-engineered answers to simple questions, and occasionally surface-level responses to genuinely complex ones. The model was guessing at the stakes from prompt length and tone, not because I'd told it anything useful about what was actually riding on the answer.

Now I can. And that changes two things at once. I stop burning depth on tasks that don't need it, and I start getting genuinely considered output on the ones that do.

The discipline underneath it is simple: match effort to stakes. A quick email reply does not need the same processing as a positioning decision that sets my pricing for the next six months. Treating them the same was never logical. Now I don't have to.

How I use the four levels

Low for things I'll barely check. Formatting, quick factual answers, proofreading. If I'm cleaning up a paragraph or turning a list into bullet points, Low gets it done faster and I don't need Claude weighing edge cases it won't find anyway.

Medium for most everyday writing and research. This is my default for routine work. A solid first draft of a newsletter section, a document summary, a quick competitor comparison. Enough depth to be useful, not so much that I'm waiting on something that didn't need the wait.

High for complex drafts and multi-step analysis. Anything feeding a bigger decision. A sales email to an important prospect, a content brief where the angle actually matters, research that informs something I'm building. These have real consequences if the output is mediocre, so I give them real effort.

Max for the 10% that genuinely matters. A pricing decision. A strategic plan. A client deliverable someone will act on. A positioning document my whole team will work from. When getting it wrong carries real cost, I stop treating it like a routine task.

2026 State of AEO Report

A year ago, most marketers weren't thinking about AI search. Now it's one of the fastest moving channels in the industry and nobody has a playbook yet.

So we built one. We surveyed hundreds of marketers to find out how they're approaching answer engine optimization, where they're investing, what's actually working, and what isn't.

The result is the 2026 State of AEO Report. Real data. Real strategies. A clear picture of where AI search is headed and how to get ahead of it.

*Ad

The Max-effort prompt I actually use

When a task is genuinely high-stakes, I don't just flip the dial. I pair it with a prompt that tells Claude what kind of thinking I want:

This is a high-stakes task and I want your maximum effort. Do not rush it.

The task: [describe the decision or deliverable in full, including context, constraints, and what's riding on it]

Before you answer:

  • Reason through multiple approaches, not just the first one

  • Consider what could go wrong with each

  • Tell me where you're confident and where you're uncertain

  • Flag any assumption that, if wrong, would change your answer

Then give me your most considered output.

Every line of that is doing a specific job. "Do not rush it" signals I want real deliberation, not a fast first draft dressed up as analysis. Describing the context and the stakes gives Claude what it needs to calibrate. Asking for multiple approaches forces it past the first plausible answer, which is often not the best one. And asking it to flag uncertainty is where 4.8's other big improvement shows up.

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

*Ad

Why the uncertainty line matters more than it looks

That "flag your uncertainty" instruction pairs with 4.8's quieter change: it's now about four times less likely to hand you a confident answer that's quietly wrong. Anthropic's internal testing scored it at 0% on uncritically accepting flawed results.

That matters because the old failure mode was never Claude giving obviously bad answers. It was Claude giving polished, confident answers with a wrong assumption buried three paragraphs in. You'd only catch it if you already knew enough to question it.

Max effort plus an explicit uncertainty request means I get real uncertainty flagging instead of polished-sounding false confidence. The combination is what makes high-stakes output trustworthy rather than just impressively formatted.

My honest take

The benchmark number is not the upgrade. The upgrade is finally being able to tell my AI how hard to think before it starts.

Here's the one habit worth picking up this week. Set effort to Max on your single most important task of the day. Set it to Low on everything routine. The difference on what matters is immediate, and the intuition compounds fast. Within a week you'll know in your gut where each level belongs, and your most important work will consistently get the depth it deserves instead of whatever the model guessed at.

Tonight, pick the one task on your plate that actually has a cost if you get it wrong. Set it to Max. Watch what comes back.