Beyond the Buzz: Is GPT-5 a disappointment, or a quiet revolution?

additional research and reporting by Soraya Panambalom, Data Science Intern

When GPT-5 was released, the internet was flooded with lukewarm reviews. Many called it a disappointment, arguing it wasn't the "giant leap" they'd been promised. So, we decided to put it to the ultimate test: a series of 16 real, professional UK accountancy exams

What we discovered is that the story isn't about one single leap. It's about a quiet revolution happening behind the scenes. The world of AI has split into two very different paths…

• The "Fast Assistant" (The Chat Models): These are the everyday AIs like GPT-4 and the standard GPT-5. They're smart, quick and getting better with every update. Our tests showed GPT-5 Chat delivered solid, consistent gains—for instance, improving the management accounting score from 78% with GPT-4o to 86%. It’s an evolution, not a revolution.
• The "Expert Specialist" (The Thinking Models): This is where the magic happens. These new "Thinking" models are designed for one thing: accuracy. They are intentionally slower because they deeply analyse problems from multiple angles. They're not just answering; they're reasoning. The difference this makes is staggering. On a difficult accountancy law exam, for example, the "Fast Assistant" scored a passable 74%. The "Expert Specialist" soared to 88%. That’s the leap from competence to true proficiency.

The First-Ever Perfect Score in our accounting evaluation

This new GPT-5 "Thinking" model is the clear leader, outperforming even its direct predecessor, the o3 thinking model. The most stunning result was on the ACCA  (Association of Chartered Certified Accountants) management accounting exam, where it achieved a perfect 100% score.

For any business that relies on accuracy, that final jump to a perfect score is everything. It's the difference between a helpful tool and a trusted advisor. It’s the moment AI becomes reliable enough for the high-stakes world of finance.

The most powerful trend we saw was a steady, undeniable march toward competence. The clearest way to see this is by looking at how many of the 16 exams the AI failed.

• 18 months ago, AI failed 6 of these exams.
• In our tests, GPT-4o failed 4.
• The new GPT-5 "Thinking" model failed only 1.

This is the real story. While the public looks for a single, flashy breakthrough, professionals should - and will - be excited by this consistent, predictable progress. For businesses, steady improvement is far more valuable than hype.

So, is GPT-5 a disappointment? If you're only using the "fast assistant," you might think so. But if you look at the "expert specialist," you'll see an AI that is quietly, but surely, mastering the complex professional world.

And for us, that's the only breakthrough that matters.