Opus 5 tops the AI ranking: what it changes for you

Every time a new model comes out, someone messages me asking if we need to "switch everything". This week it was because of Opus 5, which took first place on the Artificial Analysis AI ranking, the index that compares models on reasoning, code and knowledge tests. It is a fair question. The answer is almost always a letdown: for your business, in most cases, it changes very little. And the little that changes is quite specific.

Let me explain what that ranking measures, where a better model actually shows up in the result, and where it makes no difference at all.

What this AI ranking really measures

Artificial Analysis runs a battery of standardized tests on the models and rolls everything into a single number. Hard math problems, graduate level science questions, competitive programming challenges, multi step reasoning. The index blends those results and sorts them.

It is useful. It is also limited. Look at what is being tested: closed problems, with a right answer, with no company context whatsoever.

Your business does not work like that. Your problem is "this client replied to the quote with a slightly ambiguous question and nobody saw the message for two days". No benchmark measures that. The bottleneck there is not model intelligence. It is that nobody built the path between WhatsApp and the person who decides.

I have seen a company burn an entire meeting arguing about which model to use while the website form was sending leads to an email nobody had opened since 2023. The model was the smallest of their problems.

Switching models is easy. The hard part is knowing what you want it to do.

Where a better model actually shows up

That said, this is not empty marketing. There are three places where the jump in quality is visible to whoever uses it.

Long tasks without supervision. Before, you asked for one thing, checked it, asked for the next. Now you can hand off a ten or fifteen step task and the result comes out coherent from start to finish. That changes the kind of automation worth building. A flow that reads the spreadsheet, cross checks it against the system, spots the inconsistencies and writes the summary has become reliable. Two years ago, it broke at step four.

Bad documents. A crooked scanned contract, an invoice in PDF from a supplier who hates standards, a spreadsheet with three header rows and merged cells. Older models got these wrong exactly where you could not afford it: number, date, amount. Current ones get it right far more often. They still make mistakes, which is why you still review. But the review shifts from "check everything" to "check what it flagged as uncertain".

Replies that do not sound like a robot. Automated support stopped sounding like those phone menu trees. The model understands that "is that proposal still on the table?" and "did you ever get a chance to see my email from last week?" are the same question in different moods, and it answers accordingly.

Notice that none of those three show up properly in the ranking. The ranking measures math exams. What changes your day is reliability on long, boring tasks.

What stays the same, and will keep staying

Here is the part that matters most.

The model knows nothing about your company. It does not know that client X always pays fifteen days late and that is fine. It does not know the July promo has different rules. It does not know where the file with the updated price table lives. None of that is in the training data, and no new version is going to guess it.

Which means the work of building a good automation is the same as it always was:

• Find out exactly where the time is going
• Write down the rules that today only exist in someone's head
• Connect the systems that need to talk to each other
• Decide what the AI does on its own and what needs a human eye
• Test with real, ugly cases, not with pretty examples

A good eighty percent of a project's hours go into that. Choosing the model is a five minute decision, and a reversible one: today you switch models by changing one line of config.

So waiting for the next release to get started is the worst possible plan. You postpone the hard work because of the easy part.

A concrete example of where this weighs in

A common setup: an office gets around forty emails a day, about twenty of them repeat operational stuff. Requests for a second copy of a document, questions about deadlines, appointment confirmations. Someone spends two hours a day on that.

I built a flow for it. The automation reads the inbox, sorts messages into five categories, pulls the information from the system when needed and writes a ready reply. It does not send on its own. It leaves a draft.

In the first week, the person approved everything after reading carefully. Two hours became fifty minutes. By the third week, they were glancing and approving in batches. Twenty minutes. The odd cases, about two a day, reach them the normal way, with a note along the lines of "this one did not fit anywhere".

What made that project work was not the model. It was the decision to leave a draft instead of sending directly. That built trust, and trust is what lets you scale later. With a stronger model the classification improves a few points and the drafts get sharper. Good. But if I had let it send on its own from day one, one silly mistake would have killed the whole project, with any model in the world.

"So should I ignore these rankings?"

No. Ignore them as a sales argument, use them as a thermometer.

If you hired someone to build an automation in 2024 and it came out mediocre, it is worth reopening the conversation. Things that were unfeasible have become routine. Extraction from complicated documents is the clearest case. Flows with many chained steps too.

If you use AI in customer support and still get that "you can tell it is a bot" feeling, it is probably the old model or a poorly written instruction. Both get fixed in an afternoon.

And if you do not use AI for anything yet, the ranking tells you nothing useful today. It tells you the tool is good. You already knew that.

One practical detail: a stronger model costs more per use. Not every task needs the top of the list. Sorting emails into five categories runs perfectly on a small, cheap model. Drafting a contract analysis, that is where the expensive one earns its keep. In practice I mix the two in the same flow, and the bill at the end of the month is grateful.

What to do on Monday

If you want to take advantage of this without becoming a hostage to launch news, do one thing only: pick a repetitive task that eats more than an hour a day from someone on the team and write down on paper the exact steps that person follows.

That is it. No tools, no accounts anywhere.

If you can write down the steps, it can be automated. If you get stuck halfway because "it depends", you found the point where human judgment genuinely matters, and that is exactly where you keep the person and automate the rest.

The ranking will change again in a few months. Someone will pass Opus 5. Your internal process, that stays your problem. The good news is that it is the only one really worth solving, because it improves the result no matter which model is on top.

If you want to talk through which task in your operation is worth automating first, get in touch. No fluff, I look at your case and tell you whether it is worth it.

Every week someone asks me if they need to "switch everything" because a new model came out.

This week it was Opus 5, which took the top spot on the Artificial Analysis ranking.

The answer is a letdown: for your business, it changes very little. These rankings measure math exams and competitive coding. Your bottleneck is the lead that sat untouched for two days in an inbox nobody opens.

I have seen a company burn an entire meeting picking a model while the website form was sending contacts to a mailbox abandoned since 2023.

Switching models today means changing one line of config. The real work is finding where the time leaks, writing down the rules that only exist in someone's head, and deciding what needs a human eye.

Waiting for the next release to get started is postponing the hard part because of the easy part.

Which repetitive task in your operation eats more than an hour a day? Tell me and I will say whether it is worth automating.

#automation #artificialintelligence #productivity #digitaltransformation #ai