Was Opus 5 Really a Disappointment? Why Opinions Are So Divided
Since Opus 5 was released, reactions online have been sharply divided.
Some people say it has become dramatically smarter. Others say it feels no different from 4.8—or even worse.
It is rare to see opinions split so dramatically over an AI model.
I use Claude for work every day, and to be honest, my own impression is simple: I cannot really tell the difference.
The Problem Was Not Opus 5, but Expectations
I think the biggest reason for the backlash was that Anthropic itself set expectations too high.
Before the release, the company repeatedly emphasized that Opus 5 had surpassed Fable 5 in some areas. Naturally, users took that to mean the new model would represent a major leap forward.
But after trying it, many people seem to have reached a more restrained conclusion: it is not bad, but it is not revolutionary either.
Had it been introduced from the beginning as a model that steadily improved on 4.8, the response might have been far less contentious.
People often judge satisfaction not by performance alone, but by the gap between performance and expectation.
If you expect a score of 100 and get 85, you feel disappointed. If you expect 70 and get 85, you are impressed. The result is the same 85 in both cases.
Benchmarks and Real Work Are Different Things
Benchmarks receive a great deal of attention in the AI industry.
Scores in mathematics. Scores in coding. Improvements in reasoning.
These numbers are useful as reference points, of course. Still, I have long felt that something is missing from the discussion.
I have never used AI at work and thought, “That high benchmark score really saved me today.”
What I have often thought is, “That search was careless,” or, “That was the part I needed you to investigate.”
The quality of the actual user experience is difficult to measure with benchmarks.
Think of a car with more horsepower and better fuel efficiency. You can still get behind the wheel and feel that it has not changed very much.
AI may now be entering the same stage of maturity.
Anthropic May Have a Marketing Problem
Looking at Anthropic lately, I cannot help feeling that some of its decisions have been unfortunate.
I do not doubt its technical ability at all. In fact, it is probably among the best in the industry. But its recent marketing decisions have repeatedly worked against it.
Fable 5 was initially offered through credit-based billing, with temporary access included in subscriptions. That subscription access was extended several times before eventually becoming a permanent inclusion.
If that was where the company was going to end up, it would have been better to include it from the start. The confusion clearly damaged users’ trust.
Now we have the excessive promotion of Opus 5.
To use a baseball analogy I have mentioned before, imagine a pitcher being advertised as someone who can throw at 160 kilometers per hour. Everyone will be excited. But if the pitcher cannot throw strikes, a player who throws at 155 kilometers per hour with excellent control will be far more likely to win games.
AI is no different.
Impressive numbers matter less than having a system you can confidently rely on for everyday work.
The Real Danger Is Selling Expectations
Competition in the AI industry is intense, with new models appearing almost every week. I understand why companies feel pressure to stand out.
But long-term users are not judging the advertising. They are judging the daily experience.
Is the search accurate?
Does the model understand instructions properly?
Can it be trusted to complete the work?
In the end, evaluations always return to those questions.
I do not think Opus 5 is a failed model. But there was a gap between what many people imagined when they heard that it had “surpassed Fable 5” and what they experienced when they actually used it.
Perhaps that gap was created not by the AI, but by the marketing.
Trust built through technology takes time. Trust lost by raising expectations too high can disappear in an instant.