Artificial intelligence has been advancing at a breakneck speed, but it has also led some people to ask whether an A.I. system could go rogue and turn on its human creators. And so few companies have been as important to the rise of artificial intelligence as OpenAI. The company has tried to up the ante with each iteration of the software for what generative AI is capable of.
When OpenAI announced GPT-5, it didn’t present it in modest terms as an incremental improvement. It declared GPT-5 to be its most capable, powerful, and flexible model yet.
But it’s easy to make bold claims. The proof will be in how the technology performs in the wild. To find out, I have been testing GPT-5 for a few days. From storytelling and reasoning to programming and real-world problem-solving, I was curious to know whether GPT-5 really provides a leap forward, or whether all the hype exceeds the substance.
Here’s what I discovered.
Setting the Stage: What OpenAI Promised
When OpenAI unveiled GPT-5, it focused on a number of major enhancements:
- Sophisticated reasoning abilities — The model was supposedly adept at complex multi-step logic.
- Fewer hallucinations — A common problem in AI chatbots for a long time, GPT-5 was reported to give fewer incorrect factual answers.
- Better coding ability — Coders were informed they could depend on GPT-5 to help with increasingly advanced programming.
- More flexibility — GPT-5 could move among topics — casual conversation to detailed explanation — more easily than before, OpenAI claimed.
- Safety and alignment improvements — Each new version has aimed to be safer and more aligned with human intentions, and GPT-5 was advertised as the most reliable to date.
High bar, that. True enough, GPT-5 would not so much improve AI interaction as remake it entirely.
First Impressions: A Smoother Conversationalist
When I finally got to speak to GPT-5, I was struck by the ease with which we chatted. GPT-4 was already a good conversationalist, but GPT-5 goes even further. Its phrasing, pacing, and tone are uncannily human-like, and it appears to naturally adopt the style of the person interacting with it.
For instance:
- When I started with casual, chatty questions, GPT-5 gave warm, informal responses.
- When I shifted to technical points, it adapted quickly and gave structured, analytical answers.
The fluidity of this movement was remarkable, happening more seamlessly than I had previously witnessed.
This flexibility is key to making the model feel less like a programmed system and more like a flexible assistant.
Testing Reasoning: Where GPT-5 Surprises
OpenAI claimed that GPT-5 shows greater reasoning ability. To test it, I posed puzzles, logic problems, and scenario-based challenges.
- Travel Planning Test: I asked it to plan a three-day itinerary for a family visiting Japan, balancing culture, kids’ activities, and affordability. GPT-5 not only produced a solid plan but also explained its trade-offs. When I added a new constraint — “What if it rains on day two?” — it adjusted the schedule easily, showing situational awareness.
- Mathematical Reasoning Test: On multi-step word problems, GPT-5 outperformed GPT-4. Though not perfect, it demonstrated a clearer ability to “think through” sequences rather than jumping straight to an answer.
Of course, mistakes still surfaced — wrong calculations or misapplied rules — but these were less frequent and more easily spotted than before.
Creativity on Display
I also wanted to test GPT-5’s creative side. I asked it to:
- Write short stories
- Compose poems
- Draft marketing proposals
The results were consistently engaging.
- In a fictional science-historical drama, it generated a narrative with strong imagery and coherent structure — far more polished than earlier models that tended to ramble.
- It experimented across forms, from minimalist haiku to free verse, using tone appropriately to my instructions. The writing felt less mundane, with crisper word choices and more emotional nuance.
- For professional writing like advertising copy, GPT-5 generated persuasive, on-brand language that felt almost client-ready.
Coding: A Serious Step Forward
Perhaps the most useful leap was in coding. Developers know that AI assistants can save time but also risk introducing hidden bugs. GPT-5 improves the balance.
I tested it on Python, JavaScript, and even Rust. It not only produced working code snippets but also explained its reasoning in plain English — a valuable aid in debugging.
When given an ambiguous prompt, GPT-5 asked clarifying questions instead of making reckless assumptions. That caution felt like a sign of maturity compared to earlier versions.
Still, GPT-5 isn’t a substitute for human developers. It handled standard tasks well but isn’t ready to manage complex production code at scale without human oversight.
Tackling the Hallucination Problem
“Hallucinations” — fabricated facts presented confidently — have long plagued AI. OpenAI promised improvements here, and GPT-5 does appear more grounded.
When I asked about science, history, and niche trivia, GPT-5:
- Often provided accurate answers.
- Admitted uncertainty when it didn’t know.
That readiness to say “I don’t know” is a meaningful step toward trustworthiness.
Still, hallucinations haven’t disappeared. In one instance, GPT-5 cited a nonexistent academic paper. While such mistakes are rarer, they still demand verification.
The Human Factor: Security and Ethics
Safety was another promise. I tested GPT-5 with ethically thorny prompts.
- It frequently steered conversations toward measured, thoughtful responses.
- It handled sensitive content with care, avoiding harmful or biased replies.
That said, no AI is completely safe. GPT-5, like earlier models, inherits biases from its training data. Progress is visible, but responsibility ultimately rests with human users.
Opinion: A Leap, but Not a Revolution
After testing GPT-5 for several days, one conclusion is clear: it’s a leap forward, not a total revolution.
- It’s more fluent, adaptable, and reliable than GPT-4.
- Its reasoning is sharper, its creativity richer, and its coding skills more mature.
- Crucially, it makes fewer errors and admits limitations more openly.
Yet flaws remain:
- Hallucinations still occur.
- Complex tasks still need human oversight.
- Its conversational polish hasn’t crossed into genuine understanding.
GPT-5 should be seen as a powerful collaborator, not an infallible oracle. For professionals, creators, and casual users, it offers real-world benefits that save time and spark ideas.
Looking Ahead
The slope of AI progress remains steep. Each new system brings us closer to tools that feel more human in reasoning, invention, and collaboration.
But technology is only half the story. The future of AI will be defined not just by what models like GPT-5 can do, but by how we choose to use them responsibly.
For now, one thing is clear: GPT-5 is here, it’s powerful, and it’s worth exploring — but it still needs us as much as we need it.
