After a few days with GPT-5, I can’t share the doom-and-gloom take I’ve seen in my feed.
Yes, if you just use GPT-5 out of the box, results can be a mixed bag. But honestly, that was true for GPT-4o as well. The surprise shouldn’t be that you still need to guide it — it’s that in some areas, the leap is substantial.
Before this release, I was already using o3 for most high-quality work. Comparing o3 to GPT-5-Thinking, the improvement is obvious:
- More proactive in exploring relevant angles
- Goes beyond exactly what I ask for
- Produces more comprehensive, well-structured answers
- Shows intermediate results while still analysing — first graphs, early findings — so you can course-correct midstream
Where I notice this most is in hands-on data analysis. Give GPT-5-Thinking a dataset and ask for an exploration of correlations or trends, and the results are often junior-analyst level or better. With o3, the range was wider — sometimes that level, sometimes “first week learning Python.”
That said, the fundamentals haven’t changed:
- You still need to steer the process
- Running on autopilot is not advisable
- It’s an evolution, not a revolution — more like the steady climb from GPT-4 to GPT-4o to o-series models, not the giant leap from GPT-3.5 to GPT-4
If you go in expecting AGI, you’ll be disappointed. If you go in with a clear goal, good context, and a willingness to collaborate with the model, you might be pleasantly surprised at how far the quality bar has moved.
Curious how others are finding GPT-5 in real work — what’s better for you, and where does it still fall short?
Originally published on LinkedIn. Read the original →