GPT-5 Is Here - And It Feels Like Working With a Smarter Colleague
From Hallucinations to Honest Answers - Why GPT-5’s leap in reasoning is the upgrade business users have been waiting for.
This afternoon, I opened ChatGPT, and something felt different. Not the interface - that’s still familiar - but the way it responded. It wasn’t just giving me answers; it was thinking with me. It paused in the right places, pulled together insights that made sense in context, and when it didn’t know something, it admitted it. This wasn’t a “slightly better” ChatGPT.
This was a leap forward.
OpenAI’s release of GPT-5 today isn’t just a model upgrade. It’s a redefinition of what AI can do for business users. And in a world where trust, accuracy, and reasoning matter more than novelty, this release lands in exactly the right place.
The Big Shift: Reliability and Reasoning
One of the most frustrating parts of working with earlier AI models was their tendency to… make things up. If you’ve ever asked for a legal precedent only to discover it didn’t exist, you know the problem. GPT-5 tackles this head-on. OpenAI claims a 65% reduction in hallucinations (when using reasoning) compared to its most capable previous reasoning models, and my early testing bears that out.
In a valuation context, this matters. If you’re building an analysis for a client pitch, you need the math to check out and the citations to be real. I tried a sample scenario: “Summarize key valuation drivers for a mid-market SaaS firm, using public comps and multiples from the past 12 months.” In GPT-4, I’d often have to cross-verify every single figure. GPT-5 provided a clean, transparent summary, and when it lacked current data, it flagged that gap. That alone changes the way you can integrate AI into workflows.
Writing That Feels Like a Skilled Editor
The other shift is in writing quality. GPT-5 isn’t just assembling sentences - it’s structuring arguments and narratives in ways that feel more intentional. If you’re in a law firm drafting a client alert or in a PE firm preparing an investor memo, you don’t just need correct information. You need a coherent, persuasive structure.
I ran a test with a complex ask: “Draft a 700-word client update on a new SEC enforcement trend, balancing legal analysis with business implications.” The result wasn’t a generic legal blog - it had the rhythm of a seasoned associate who knows their audience. The intro hooked me, the body was logically organized, and it wove in a high-level risk framework without burying the reader in citations. That kind of adaptive writing is where GPT-5 feels less like a tool and more like a team member.
Analytics That Can Actually Think
The new reasoning engine doesn’t just mean fewer hallucinations; it means better analytics. In legal due diligence, for instance, you might feed the model excerpts from multiple agreements and ask for patterns in change-of-control clauses. In valuation work, you might drop in financial summaries from several targets and ask for thematic trends in margin profiles.
In one test, I gave GPT-5 three years of anonymized operating data for a portfolio company and asked it to highlight both the key drivers of revenue growth and the operational risks the data hinted at. GPT-4 might have spit out a laundry list. GPT-5 grouped the drivers into strategic themes, connected them to market dynamics, and - importantly - explained why those risks mattered in an investor context. It wasn’t doing math for math’s sake; it was reasoning in a way that shaped the business conversation.
When You Need It to Code, It Delivers
While my focus here is on analytics and writing, I’d be remiss not to mention what GPT-5 can do in a coding context. For teams that occasionally need a quick proof-of-concept app or a custom data visualization, this model can produce production-grade code with far less back-and-forth.
The new 400,000-token context window (in the API - the web interface is 128,000 for pro; 32,000 for plus) is a big deal here. It means you can drop in an entire codebase - or a complex legal document repository - and have the model work across the whole thing without breaking its train of thought.
In a brief test, I had it create a custom Chrome plug-in to retrieve market data via an API, format it, and feed it into a valuation model template. The code ran clean on the first try. That’s time back to focus on the analysis, not the debugging.
Why This Matters for Business Users
In my world, GPT-5 is the first AI release that could have a standing seat at the table for day-to-day business work, not just a utility you turn to for quick lookups.
For law firms: More reliable research summaries, better first drafts of client communications, and the ability to spot thematic legal risks in large document sets.
For private equity and valuation firms: More accurate data synthesis, structured reasoning for investment memos, and automated coding for portfolio analytics tools.
We’ve talked for years about AI as a “co-pilot.” GPT-5 doesn’t feel like a co-pilot anymore. It feels like a colleague - the kind who comes to the meeting already briefed, with a few smart ideas on the whiteboard, and the awareness to say, “I’ll need to confirm that detail” instead of bluffing.
Looking Ahead
No, GPT-5 isn’t perfect. You still need to check the outputs. But for the first time, I’m finding I can trust its reasoning enough to start in the middle of the conversation, not at the beginning. That’s a big shift in how I’ll use it in practice - and for business users who live and die by the quality of their analysis, writing, and judgment, it might just be the difference between “AI as an experiment” and “AI as part of the firm.”
Thanks for reading. If you enjoyed this and are not a subscriber, please subscribe. If you found this interesting, please share it with your friends.
If you have a moment, please take a look at my updated website at https://intelligencebyintent.com/ - I would love to hear your thoughts/feedback!