The launch everyone was waiting for
OpenAI launched ChatGPT-5 with more fanfare than usual. Live events, demos, benchmark comparisons. The hype was enormous.
I've been using it daily as part of my developer workflow for three weeks. Here's my honest review, without the filter of initial excitement.
What actually improved
Reasoning on complex tasks
This is real. ChatGPT-5 reasons better on problems that require multiple steps. When I ask it to analyze a software architecture with cross-dependencies, the responses are more coherent and less likely to contradict themselves halfway through the analysis.
For complex debugging, I notice a genuine difference. I paste a complicated stack trace with context, and the hypotheses it generates are more accurate. GPT-4 sometimes got lost in symptoms. GPT-5 gets to the root cause more often.
Consistency in long conversations
GPT-4 had a known problem: as conversations extended, it would start "forgetting" initial context and responses would lose coherence. GPT-5 handles long conversations much better. You can have a 40-message exchange and the model stays aligned with the original context.
Cleaner code out of the box
The code GPT-5 generates needs less editing than its predecessor. It follows the patterns in your provided context more consistently, and the common logic errors that appeared in GPT-4 are less frequent.
What stayed the same (or close to it)
Hallucinations didn't disappear
It still makes things up. Less than GPT-4, yes. But if you ask about documentation for a specific library or API versions, always verify. Blindly trusting any language model in 2026 is still a mistake.
Context limits are still a practical problem
Even though it technically has a larger context window, in practice when you feed it large codebases, response quality drops toward the end of the window. Not as dramatic as earlier versions, but it's there.
Slowness in reasoning mode
The extended reasoning mode — similar to the "thinking" they introduced earlier — is slower. I get the tradeoff, but when you want a quick answer, waiting 30 seconds of "reasoning" gets frustrating.
Is it worth the price?
ChatGPT Plus went up to $25/month with GPT-5 available. For casual use: probably not. The free version with limited GPT-5 access covers most use cases for someone who doesn't use it intensively.
For professionals who use it as a work tool: yes, the price is justifiable if you were already paying for GPT-4. The improvement in reasoning and code quality is enough to justify the difference.
My personal use: I have it integrated in Cursor as a model provider and in a personal workflow I described in another post. In that context, GPT-5 is a real upgrade over GPT-4 Turbo.
The comparison nobody wants to make
Anthropic's Claude 4 Opus remains competitive in reasoning and code. In some specific tasks — especially analyzing extensive code — I still prefer Claude. The gap isn't as wide as it was a year ago, but it exists.
No single model wins at everything. In 2026 the smart strategy is having access to more than one and knowing when to use which.
Verdict
ChatGPT-5 is a real improvement, not empty hype. But it's also not the revolutionary leap the marketing wanted you to believe. It's a solid evolution that improves what was already good and patches some of the most obvious weaknesses.
If you were already using ChatGPT for serious work: upgrade. If you were casual: the free version with limited access is enough. If you've never used any model: start with the free version and evaluate whether your usage justifies paying.