Big Idea

The argument in one line.

Opus 5 needs the opposite of old step-by-step prompting: hand it the whole job at once, then rein in its tendency to overdeliver with explicit limits on scope, response length, and project size.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You're on a paid Claude plan (Pro $20 or Max $100+) and use Claude for real knowledge work, not just casual questions.
You've been prompting Claude the same way for the past year and want to know exactly what changes with Opus 5.
You want a concrete, reusable prompt template rather than abstract prompting theory.

SKIP IF…

You don't have access to Opus 5 or only use Claude for occasional chat questions.
You're looking for Claude Code engineering workflows specifically — this covers general knowledge-work prompts like websites, docs, and plans.

TL;DR

The full version, fast.

Claude Opus 5 is Anthropic's strongest Opus model yet, but prompting it like earlier Claude versions leaves results on the table. Model choice depends on plan: Sonnet by default on the $20 plan (switching to Opus 5 for hard tasks), Opus 5 as the default on $100-plus Max plans, with Fable 5 reserved for comparison when Opus 5 disappoints. Effort level controls thinking time and token spend, not automatic quality, so low or medium is often enough. Anthropic's own prompting guide gives five rules: hand over the entire job upfront instead of breaking it into steps, set explicit limits on what not to do, specify how short the response should be, cap the size of the actual work product, and stop telling it to double-check its work, since Opus 5 already does that as it goes.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:44

01 · Intro

Opus 5 is positioned as the best Opus model yet; Anthropic published its own prompting guide because old habits underperform on it.

00:44 – 02:36

02 · Which Claude Model Should You Use?

Model choice by plan: skip Haiku for real work, Sonnet-default on Pro, Opus-default on Max, Fable 5 only to compare.

02:36 – 03:40

03 · Choosing the Right Effort Level

Effort controls thinking time and tokens, not guaranteed quality; low/medium is often enough.

03:40 – 04:14

04 · Building the Prompt

Sets up the Northline Coffee case study used to demonstrate all five rules together.

04:14 – 05:21

05 · Rule 1: Give the whole job upfront

State the full goal, assets, and end point in one prompt instead of stepping through it.

05:21 – 06:04

06 · Rule 2: Set clear limits

Opus 5 overdelivers by default; tell it what to leave alone and when to ask before proceeding.

06:04 – 07:59

07 · Rules 3 and 4: Constrain the response and the work size

Cap both the length of Claude's text reply and the size of the actual deliverable — two separate limits.

07:59 – 09:18

08 · Rule 5: Don't make it check the work twice

Drop the old 'verify everything twice' instruction — Opus 5 already checks and fixes as it goes.

09:18 – 12:18

09 · Testing the Finished Prompt

The five-rule prompt is run live in Claude Design; the site finishes in under three minutes, matching every constraint.

Atomic Insights

Lines worth screenshotting.

Opus 5 sometimes beats Fable 5 on benchmarks despite being the cheaper, less resource-intensive model.
On the $20 Claude plan, Sonnet should be the default model, with Opus 5 reserved for tasks Sonnet can't handle.
On $100-plus Max plans, Opus 5 can be the default for nearly all knowledge work, with Fable 5 needed only to compare against a disappointing result.
Effort level controls how much thinking time and how many tokens Claude spends, not whether the output quality improves — higher effort is often just overkill.
Anthropic recommends low effort on the Pro plan and medium effort as the default on Max plans for normal knowledge work.
Older Claude models rewarded breaking a task into small sequential steps; Opus 5 performs better when given the entire job in one prompt.
Opus 5 tends to do more work than requested, so prompts need explicit limits on what NOT to build or change.
Telling Opus 5 to check its work twice can waste time and tokens without improving the result, because it already checks and fixes its work as it goes.
Constraining response length (a fixed number of bullet points) is a separate lever from constraining the size of the actual deliverable — both need to be specified.
A precisely scoped prompt let Claude Design build a complete six-section coffee brand website in under three minutes without asking a single follow-up question.

Takeaway

Five rules for prompting Claude Opus 5 the opposite way

WHAT TO LEARN

Opus 5 rewards handing over the whole job at once and reining in its tendency to overdeliver, rather than the incremental step-by-step prompting that worked before.

01Intro

Claude Opus 5 is being positioned as Anthropic's best Opus model yet, occasionally beating even Fable 5 on benchmarks.
Anthropic published its own prompting guide specifically because pre-Opus-5 prompting habits produce worse results with the new model.

02Which Claude Model Should You Use?

Haiku is fine for casual chat questions but isn't worth using for any real knowledge work.
On the $20 Claude plan, default to Sonnet and switch to Opus 5 only for complex tasks or when Sonnet's results fall short.
On $100-plus Max plans, Opus 5 can be the default for almost everything, cutting how often Fable 5 is needed.
Fable 5 is now mainly useful as a second opinion when you're unhappy with what Opus 5 produced, not as a default upgrade.

03Choosing the Right Effort Level

Effort level controls how much time and tokens Claude spends thinking, not a guaranteed quality boost — higher effort is often unnecessary overkill.
Anthropic's own guidance says low and medium effort are surprisingly strong for normal knowledge work.
A simple default — low effort on Pro, medium on Max — beats manually switching effort per task for most people.

04Building the Prompt

The video builds one real prompt end-to-end for a fictional coffee brand to demonstrate all five rules together rather than in the abstract.

05Rule 1: Give the whole job upfront

Older Claude models rewarded splitting work into small sequential steps; Opus 5 performs better when handed the entire job, context, and end goal in one prompt.
A well-scoped 'full job' prompt states the goal, the assets available, and where the finished work should end up, all in the first message.

06Rule 2: Set clear limits

Opus 5's tendency to do more than asked means prompts need explicit boundaries: what to build, what to leave alone, and when to ask before proceeding.
Telling Claude to make small design decisions itself, but only surface questions that would change the whole project, keeps it moving without losing control.

07Rules 3 and 4: Constrain the response and the work size

Rule 3: separately constrain how long Claude's text response should be (e.g., three short bullets) so you actually read the summary instead of skipping a wall of text.
Rule 4: constrain the size of the actual deliverable itself (fixed number of sections, word counts per section) — a short response and a short output are two different limits.

08Rule 5: Don't make it check the work twice

The old habit of telling AI to check every file twice and find every mistake no longer helps with Opus 5 — Anthropic says it already checks and fixes its own work as it goes.
Adding a redundant verification pass mostly just burns more time and tokens without improving the final result.

09Testing the Finished Prompt

The five-rule prompt, plus uploaded brand assets, was pasted into Claude Design on Opus 5 at medium effort — the same rules apply across chat, Cowork, the desktop app, and Claude Design.
Because the prompt front-loaded every constraint, Claude Design didn't need to ask clarifying questions and finished the full site in under three minutes.
The output matched the prompt's constraints exactly: six sections as specified, and a three-bullet summary instead of a long explanation.

Glossary

Terms worth knowing.

Effort level: A Claude setting (low/medium/high/extra) that controls how much internal reasoning time and how many tokens a model spends on a task before answering — separate from which model you pick.
Claude Design: A Claude tool for generating visual work like websites and presentations from a single natural-language prompt plus uploaded brand assets.
Claude Cowork: Anthropic's task-assistant surface inside Claude for delegating recurring work like inbox triage or daily briefings; the same prompting rules apply there too.

Resources

Things they pointed at.

00:20linkAnthropic's official Opus 5 prompting guide ↗

09:09toolClaude Design

Quotables

Lines you could clip.

00:20

“Anthropic agrees with this. So they published an entire prompting guide for Opus five.”

sets up the whole video's premise in one line→ TikTok hook↗ Tweet quote

04:10

“The first rule is that Opus five works best when you give it the full job at the start.”

the single most actionable line in the video→ IG reel cold open↗ Tweet quote

08:18

“Anthropic says that Opus five already checks and fixes its work as it goes along.”

counterintuitive, reverses years-old prompting advice→ newsletter pull-quote↗ Tweet quote

11:39

“Let Opus five use its intelligence without letting it overuse its intelligence.”

clean closing thesis, quotable wordplay→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Claude Opus five is here, and on paper, it is the best Opus model Anthropic has ever released, sometimes even better than Fable five. But if you prompt it the same way you've been prompting previous Claude models, you are not going to get good results with it.

Anthropic agrees with this. So they published an entire prompting guide for Opus five. It is dense, but I read through this entire thing so you don't have to.

And I pulled out five rules that change how I'm going to use Opus five for knowledge work. And some of the rules are the exact opposite of the advice we've been following for years. Let me break down for you what makes Opus five so unique, when you should be using it, and with what effort level, and then we'll go through those five prompting rules for Opus five.

This all starts with choosing the right model and effort level, which is more confusing than ever since we now have four models from Anthropic. The weakest model is Haiku, and the most powerful model is Fable five.

Although on many benchmarks, Opus five actually beats Fable five.

Regardless, Fable five is the most expensive. So the right model really depends on your use case and which plan you have. For myself, I never use Haiku.

You could maybe use it for general questions if you're just using Claude for chat. But for any work, I really wouldn't use Haiku at all.

Next, if you're using Claude for knowledge work and you're on the $20 Claude plan, you should use Sonnet as your default model and then switch to OPUS five for your more complex tasks or when Sonnet isn't giving you the results that you're happy with.

But if you're on the max $100 plan or higher and you're using Claude for knowledge work, you can often get away with using Opus as your default model.

In the past, I would have suggested that you switch to Fable for more complex work, but I suggest that even less now with the new Opus five, it's just that good. The only time I would switch to Fable five is when you're unhappy with the results that you're getting with Opus five and wanna compare it against the results that you get with Fable.

But less and less, I don't think it's necessary to Fable, and you can stick with Opus five for most of your tasks. You can make it more complicated by switching between Sonnet, Opus, and Fable depending on the task.

That would be more optimized, but I like to keep it simple. So now I just leave it on Opus five almost all the time.

Okay. So let's assume you're going to use Opus five. The next decision is what effort level are you going to use.

And this is where I think a lot of people are going to make a mistake. They just assume that higher effort levels mean that it's going to produce better results, but that is not the way that Anthropic describes it.

The effort setting mainly controls how much thinking Claude does, which affects the time and tokens it spends on the task. While that can produce better answers, sometimes, a lot of the time, it's just overkill.

Anthropic has said that low and medium effort levels are surprisingly strong on Opus five. For a lot of normal knowledge work, I would start there. Use low if you have the pro plan and use medium as a default if you are on one of the max plans.

Again, you can make it more complicated, but I prefer having a simple default that I use every time, which for me now is Opus five on Medium.

Once you decide to use Opus five and you pick the effort level, there are five rules I would follow when writing the prompt. Okay. So as I run through these five prompting rules, we are going to build a prompt together.

So let's assume we have a company called Northline Coffee, and we want help with building a website for this brand.

So as we kind of go through the prompting suggestions, we will build on this prompt. But as a little bit of background, this is a small coffee brand selling beans online.

We have some logos and colors and product images already that we are going to give Opus, and then we want the website to actually sell more coffee for us. So the first rule is that Opus five works best when you give it the full job at the start.

A lot of these older models, if you had a task for it, you would often break it down into smaller steps. You might say, first, look at these files and pull from it the branding for my company.

Then I want you to put together a brand guide. And then after that, I want you to put together a plan for what the website will look like. And you can still do that, but Opus five is better at taking a full job and working through it on its own.

So we're not going to start with something like look at these images and create a brand guide for us. We're gonna start with everything.

So I will tell it, build a one page website for Northline Coffee, a small brand that sells coffee beans online. The goal is to help people understand the brand, choose a coffee, and place their first order. Use a logo, colors, product details, and photos in the brand asset folder.

Build the finished website in the project folder. And already that is so much better because it gives the whole job up front where we are starting and where we need to end up. The second rule fixes a common issue with Opus five.

It loves to do more work than you ask it to do. That can be helpful sometimes, but often it results in Claude going overboard and doing more than you really need or wanted to do. So tell it when the job ends and add some clear limits.

For our prompt, we will add this. Build only the home page. Link the buy buttons to the shop page that already exists.

Keep the current logo, colors, and product name. Make small design choices yourself. Only ask me a question if the answer would change the whole website.

So now Claude has room to work, but it also knows what not to do. The third rule also fixes an issue with Opus five, which is it loves to give you long answers when it finishes. Too long.

Oftentimes, it's so long I don't even bother reading it. If that bothers you as much as it bothers me, tell it to give you a shorter answer.

For this job, I don't need a long story about everything Claude did when it created the website for us. I only need to know that the website is ready and where to find it. So we will add this.

When you finish, tell me where the website is and give me no more than three short bullets explaining what you made. Then tell me only what's truly important to know about how the website works. That will keep the response at a readable length and only give me the information I actually need.

Similarly, you also wanna limit the size of the work product, which is different from the text response. So if we left the prompt as is, we might get a short response from Claude, but the website could be too large. So we also need to set the size of the thing it's making.

For a website, we will add this. Keep the website to six parts. The top section, three featured coffees, why Northline is different, customer reviews, common questions, and one final buy button.

Keep the writing short. The main headline should be eight words or less. Each part should be easy to scan and should have no more than 70 words.

Now you don't have to be this precise. You could say something like, I only want the website to have six parts and that's it and leave it up to Opus five's discretion as to what those parts are.

Or you can say something like, I want someone to be able to read my entire website in under a minute and a half. So make sure it's no longer than necessary to convey the points that we wanna make.

The point is you want to give it some guidance on the size of the project that you wanted to create. That will get you closer to the results you want faster.

This next prompting tip isn't about what you should tell Opus five, it's what you shouldn't tell it. For a while now, we've been telling AI to check the work that it's done.

After it's done, check every file twice, find every possible mistake, and fix all of them. And usually, that would get you better results.

However, Anthropic says that Opus five already checks and fixes its work as it goes along. If we force it to do the whole check again, it may take more time and use more tokens without actually making the end result any better.

So we will not be adding that, and I suggest you don't add that to your prompts now when you use Opus five. We will still tell Claude what good looks like. We just don't have to have it repeat the same check over and over again.

So here's the full prompt now. It builds on all five of the rules that we just went over together and gives Opus five the freedom to be creative but within a set of boundaries that make sense for the project.

We gave it enough without giving it too much, and you can use these prompt rules anywhere you access Opus five. So whether that's chat on the web or co work on the web or from your phone or the Claw desktop app or even inside of Claw design. That's what I'm going to use for this design.

So I'll paste that prompt into the prompt box, and on the model selector, make sure I have Opus five selected, and the effort level, I will leave on medium. Now I need to attach the brand assets here. These are all the images of the coffee and the logos as well.

So let me do that. And now I can send that along. Usually, Claude has some questions for me when I use Claude design about what this should look like.

But sometimes if I give it enough information like I've done here, it may not ask me those follow-up questions. Let's see what it does this time. I guess it's just getting to work now.

And we can kinda see it building it from the top down here. Okay. So it's now done now.

That took a little less than three minutes, which is quite fast for Claw Design. Usually, when I have it create websites for me, it takes a lot longer. But I think because we were so precise with our prompt and we gave it exactly what we're looking for, it was able to produce something much faster because it knew what we were going for right from the start.

I mean, like we were saying, tell it what you want. Give it the entire project. And I think that's also why it didn't ask me any follow-up questions because it already knew.

It was all in the prompt. And, otherwise, it looks really nice. Like, it gave it enough freedom to have some creativity to make it look nice.

It brought in the images that I uploaded. So overall, I'm really happy with this. This is exactly what we were looking for, and it was done quite quickly.

But let's count this up. We have section one, two, three, four, five, six, which matches exactly what we asked for.

And each of these sections lines up. Right? We have our hero section here.

We have the products here. We have a little bit of information about Northline, some reviews, some common questions, and then a call to action down here. Then coming over to the chat, what do we ask for?

We ask for three bullet points that explain what was done, and that's all we got. Three bullet points.

So really nice that it actually followed those directions instead of just giving us this big wall of text that we would never read. Now we can actually read it and we're not gonna, you know, dismiss it because it's so long. So that's how to prompt Opus five the right way.

Once you choose Opus five, start with an effort level that matches the task and the plan you're on, then follow the five rules that I gave you. In other words, let OPUS five use its intelligence without letting it overuse its intelligence. But the model is only half the story.

You also have to know how to use the tools, whether that's Claude chat, Claude Cowork, or Claude design. So if you want a full breakdown on how to use the new Claude, including how to use Cowork on the web and on your computer, check out this video right here. I made this very recently, so it has all the newest Claude updates.

So if you want the newest, most up to date video on how to use Claude, definitely click on this video right here, and I'll see you over there in one second. Bye.

The Hook

The bait, then the rug-pull.

Anthropic just shipped Claude Opus 5 and quietly changed how it wants to be prompted. This breakdown pulls five rules from Anthropic's own prompting guide, then tests all five live by building a real one-page website from a single prompt.

Frameworks